Professional Documents
Culture Documents
Umts PDF
Umts PDF
UMTS Technolog
Advances in
UMTS Technology
edited by
JC Bic & E Bonek
Tontents
Foreword
J. C. Bic, E. Bonek
1. Third generation mobile systems UMTS/IMT-2000
J.-P. Charles
VII
1
12
61
92
126
146
177
8. Wireless communications + + +
R. Steele
196
Index
213
Foreword
In recent years enormous research effort has been devoted all over the world
to specify, create and develop efficient radio interfaces and access network
architectures in order to provide new services. Research laboratories, mobile
operators, manufacturers, regulators have all contributed to the definition of
a world-wide system. This so-called third generation mobile system is now
coming to reality in Europe and Japan by the name UMTS (Universal Mobile
Telecommunication System). The main features of UMTS are now well known:
Spectrum efficient radio interfaces based on spread-spectrum and CDMA
techniques, and sophisticated modulation and coding methods offering
high capacity.
Large bandwidth enabling broadband services with bit rates several times
larger than enhanced second-generation systems, even if the 2 Mbit/sec bit
rate per user would likely be limited to picocells.
Ability to interconnect with IP-based networks, paving the way to truly
fixed-mobile networks convergence.
Flexibility of mixed services with variable data rates, providing a wide
range of services from low-rate speech to interactive multi-media
communications.
Now one of the most exciting challenge for the coming years is the deployment of these complex networks both from technical and financial viewpoints.
Even if the planning is not so optimistic as it was one year ago, operations will
certainly begin in 2002.
New services are crucial for the success of UMTS. Although the general
service principles are stated (Open Service Architecture), the "killer application"
is still well kept in the drawers of operators and manufacturers, and that is why
this aspect is not deeply investigated in this publication.
In parallel with the implementation of the standards, research especially on
the air interface is still proceeding at a rapid pace for even better capacity,
quality and flexibility with enhanced transmitters/receivers.
This publication will address several issues related to UMTS emphasizing
future evolution to improve the performance of Third-Generation Wireless
Mobiles on the way to Fourth Generation. The contributions come from
academic scientists, manufacturers and operators.
VIII Foreword
Foreword
IX
Chapter 1
I. Introduction
With third generation mobile systems, the world of mobiles will enter the era of
multimedia. The stakes are considerable: around 2010, mobile traffic should be
equal to that of fixed telephony. The convergence of mobile and Internet worlds,
the strong dynamics of innovation, and the reduction of costs in these domains
will open new opportunities for multimedia services. These systems could be
brought into service as early as October 2001 in Japan, and around 2002 in
Europe, in new frequency bands around 2 GHz.
Several specific applications will use the capacities of UMTS systems to provide
data, images or even videos: video medical diagnosis, reporting, proximity
services, remote control, information, and driving guidance. Professionals' needs
will also be satisfied through access to different means of telecommunication.
UMTS will provide true mobile offices, even in vehicles.
Beyond professional use, the reduction of costs will lead to the generalization
of these personal multimedia tools, the use of which should gradually extend to a
large customer base following mobile telephony. Young people will spur the
development of this market through their needs for games, education, sports.
Thus, by the end of 2004, according to a number of studies, there will be
120 million multimedia mobiles out of a total of 1.1 billion subscribers in the
world, and 4 out of 10 Internet users will also use mobile access to Internet at that
time. Concerning data rates, UMTS is expected to offer up to 2 Mbit/sec, whereas
GSM/GPRS can only support around 100 kbit/sec.
a completely new radio access network. Several thousand UMTS base stations
will be needed to offer national coverage with high data rates and reuse of
existing GSM radio sites will be a key issue to deploy rapidly.
The adoption by ETSI of FDD/W-CDMA opened the doors for an agreement with
Japan, about to adopt this technology for its own third generation mobile system.
Discussions among various standardization organizations: ETSI for Europe, TTC
and ARIB for Japan, TTA for Korea, T1 for the United States, led to the creation, in
December 1998, of a partnership among these organizations called 3GPP (third
generation partnership project). This forum (http://www.3gpp.org) developed the
technical specifications for UMTS. Then, these specifications were adopted as
standards by the different national or regional standardization bodies.
10-5 - 10-8
Multiple access
Bit rate
Carrier spacing
Frame length
Frame structure
Modulation
Spreading factor
Channel coding
FDD
TDD
(Frequency domain
duplex)
DS-CDMA
TDMA/CDMA
3.84 Mchip/s
4.4 to 5 MHz with a 200 kHz raster
10 ms
15 time slots per frame
QPSK
4 to 256
1 to 16
Convolutional (rate 1/2 to 1/3)
Turbo codes for BER < 10-3
10
VI. Conclusion
The choice of the principles of the UMTS radio interface in January 1998 gave a
strong acceleration to the standardization process throughout the world. This
decision was particularly important, because it consolidated the technical
agreement between Japan and Europe on the adoption of CDMA as a common
11
basis for UMTS. However, this was only the first step leading to the launch of
UMTS networks in October 2001 in Japan and in 2002 for Europe. In 2000, most
of the European countries have allocated UMTS licences using beauty contest or
auctions procedures, to give sufficient time to the UMTS operators to prepare the
launch of their services in 2002. A first release of the UMTS standard which is
called release 99, was adopted at the beginning of 2000, and this release will be
used by the manufacturers for the first generation of UMTS equipment. The
competition between operators will mainly be based on their ability to provide to
their customers new services because, when UMTS is launched, a high percentage
of the population will have a mobile for telephony and it will be very difficult,
especially for a new entrant, to attract new customers with existing services. The
key aspect of UMTS will be access to high data rates and multimedia services for
the customer and, without such services, it will be difficult to transform this
costly adventure into success.
REFERENCES
[1] HOLMA (H.), TOSKALA (A.), wcDMA for UMTS, John Wiley & Sons, (2000).
[2] MOULY (M.), PAUTET (M-B.), The GSM system for Mobile Communications, (1992).
[3] BLANC (P.), CHARBONNIER (A.), VERRIER (D.), L'UMTS: la generation des mobiles
multimdia, L.'cho des recherches, n 170, 1er trimestre, (1998).
[4] HAARDT (M.), KLEIN (A.), KOELHN (R.), OESTREICH (S.), PURAT (M.), SOMMER (V.),
ULRICH (T.), The TD-CDMA based UTRA TDD mode, IEEE Journal on Selected Areas
in Communications, 18, n 8, pp. 1375-1384, (Aug. 2000).
[5] Acx (A.G.), MENDRIBIL (P.), Capacity evaluation of the UTRA FDD and TDD modes,
49th Vehicular Technology Conference, Houston, 3, pp. 1999-2003, (1999).
[6] FRANCE TELECOM, Technical analysis and comparison of UTRA concepts, ETSI SMG2
Adhoc n 4, Tdoc SMG2/UMTS 126/97, Helsinki, (17-21 Nov. 1997).
Chapter 2
Improvements in W-CDMA:
principles and experimental
results
M. Sawahashi, K. Higuchi, and S. Tanaka
Wireless Research Laboratories, Japan
F. Adachi
Graduate School of Engineering, Tohoku University, Japan
I. Introduction
Associated with the successful planned introduction of global commercial
wideband code division multiple access (W-CDMA) [1], [2] service from this year,
the dawn of the genuine era of wireless Internet is upon us. The achievable
maximum information bit rate guaranteed by the required quality level in the
IMT-2000 is 2 Mbps and in the near future the peak bit rate of nearly 10 Mbps
will be possible for high-speed downlink packet access (HSDPA), which is now
undergoing standardization in the Third Generation Partnership Project (3GPP).
Therefore, rich services such as Internet access and the transmission of video
and high-quality images from/to moving vehicles will be achieved in the wCDMA system. DS-CDMA wireless access, on which W-CDMA is based, has
numerous advantages over TDMA or FDMA including single frequency reuse, soft
hand-off (or site diversity), enhanced radio transmission through Rake
combining, and direct capacity increase through sectored antennas. The key
features of the W-CDMA physical layer are:
- Inter-cell asynchronous operation and three-step fast cell search
- Flexible realization of various levels of quality of service (QoS) for various
transport channels by rate matching associated with channel coding
- Signal-to-interference power ratio (SIR)-based fast transmit power control
(TPC) for satisfying the required quality level for a physical channel with
minimum transmit power
13
- Significant gains in link capacity and coverage through the use of many
diversity techniques, e.g., coherent Rake time diversity using pilot symbol
assisted (PSA) channel estimation, space diversity, inter-cell (sector)
diversity, and transmit diversity (only in the forward link)
- High flexibility in offering different multirate services (up to 2 Mbps)
through orthogonal variable spreading factor (OVSF) multiplexing and
orthogonal multicode transmission
- Capacity enhanced techniques such as interference cancellation (IC) and
adaptive antenna array diversity (AAAD).
The above essential W-CDMA technologies associated with its performance and
the features of the W-CDMA air-interface were comprehensively overviewed in [13]. However, in the ongoing worldwide standardization process in the 3GPP, the
radio link parameters and channel structure have been modified, and enhanced
techniques such as turbo coding for high-rate data transmission and transmit
diversity were adopted into the standards. Therefore, this chapter overviews the wCDMA enhanced wireless access technologies including the channel structure and
spreading code assignment in the physical layer and transport channel multiplexing
into a physical channel associated with rate matching and reports on a series of
laboratory and field experiments conducted in an area near Tokyo. We designed and
developed an experimental system comprising a coherent multistage interference
canceller (COMSIC), coherent adaptive antenna array diversity (CAAAD) receiver in
the reverse link, and adaptive antenna array transmit diversity (AAA-TD) in the
forward link in order to demonstrate the suppression effect on multiple access
interference (MAI) and multipath interference (MPI). The experimental results of
these techniques are also presented.
14
Improvements in W-CDMA
value of 0.667 msec taking into account a tradeoff between frame efficiency and
tracking ability of fast TPC and channel estimation against fast fading variation. The
number of channel-coded information bits, which each physical channel conveys,
differs according to the type of physical channel and spreading factor (SF). The
features of the major physical channels are described below.
(1) P-CCPCH (Primary-Common Control Physical Channel)
One P-CCPCH is defined for each sector in the forward link. The P-CCPCH has
a fixed SF of 256 (15 ksps) and carries the BCH transport channel. It is not
transmitted during the first 256-chip duration, but instead the P-SCH and S-SCH are
transmitted during that period at each slot.
(2) S-CCPCH (Secondary-Common Control Physical Channel)
Multiple S-CCPCHS, which are common channels in the forward link, are
defined in each cell (sector) and carry paging information and lower data
information from a higher layer.
(3) PRACH (Physical Random Access Channel)
Multiple PRACHS, which are common channels in the reverse link, are defined
and used to carry the RACH transport channel comprising lower information data
from a higher layer.
(4) DPCH (Dedicated Physical Channel)
A DPCH is assigned to each mobile station (MS) in both the forward and
reverse links. It comprises a DPCCH (Dedicated Physical Control Channel) and a
DPDCH (Dedicated Physical Data Channel).
A DPDCH consists of a channel-coded data sequence and more than one DPDCH
can be assigned to one DPCH. A DPCCH is used for Layer 1 control of DPCH and
one DPCCH is defined for one DPCH. A DPCCH comprises pilot bits for coherent
channel estimation, TPC bits, TFCI (Transport Format Combination Indicator)
bits, and FBI (Feedback information) bits designating the control information for
transmit diversity in the forward link (thus, FBI bits are defined only in the
reverse link).
(5) CPICH (Common Pilot Channel)
A CPICH is the common pilot channel used for channel estimation, path search
for Rake combining (generation of power delay profile), and the third step, i.e.,
scrambling code identification in the three-step cell search method. Two kinds of
CPICHS are defined: primary-CPICH and secondary CPICH. The primary-CPICH has
two-symbol data sequences associated with two antennas. Without transmit
diversity all symbol sequences with all "1" are transmitted from Antenna #1, and
with transmit diversity, the second primary-CPICH with different symbol
sequences from those of the first primary-CPICH are also transmitted from Antenna
#2 in addition to the first primary-CPICH.
15
Figure 1. Frame structure of DPCH (a) reverse link (b) forward link.
16
Improvements in W-CDMA
The frame structure of the DPCH in the reverse and forward links is
illustrated in Figures l(a) and l(b), respectively. The DPDCH and DPCCH are codemultiplexed into I and Q channels, respectively, in the reverse link. Since the
DPCCH with a fixed rate (SF) and DPDCH with variable date transmission are
separated from each other in the orthogonal phase, fluctuation of the amplitude
during variable transmission can be decreased. Meanwhile, the DPCCH and DPDH
are alternatively time-multiplexed within a slot in the forward link.
Table I. Spreading code assignment.ent
Forward link
Channelization code
Scrambling code
CPICH
#0
SF = 256
P-CCPCH
#l
SF = 256
Primary
S-CCPCH
Arbitrary
SF = 4-256
Primary (Secondary)
Primary
DPCH
Arbitrary
SF = 4-256
Primary (Secondary)
AICH
Arbitrary
SF = 256
Primary (Secondary)
PICH
Arbitrary
SF = 256
Reverse link
| DPCH
Code-channel identification in
multicode transmission (4-256 chips)
Arbitrary
SF = 4-256
Primary (Secondary)
User identification (38,400 chips)
Primary (Secondary)
17
The k OVSF codes of the k-th layer are orthogonal to each other. Furthermore,
any two codes belonging to different layers are orthogonal except for when one
code is not the mother code of the other. For example, Cch,2,0 and Cch,4,2 are
orthogonal to each other. When Cch,2,0 is already assigned, any code below this
code on the code tree cannot be used, this is a restriction of the code assignment.
The codes of Cch,256,0 and Cch,256,1 are commonly used for all cells for the
P-CPICH and P-CCPCH in the forward link, respectively. The channelization codes
of other physical channels are assigned from a higher layer.
18
Improvements in W-CDMA
where
II.2.4. Spreading
In the reverse kink, the channelization code is independently spread into I/Q
channels by using different OVSH codes and weighted by weighting factor G,
which denotes the transmitted amplitude (power) ratio of DPCCH to DPDCH.
Complex spreading is applied to the physical channel: one is a code truncated by
38,400 chips from the beginning of the Gold sequence with the repetition period
of 224, and the other is truncated by 38,400 chips of the shifted first Gold
sequence by 16,777,233 chips. Thus, the spreading using channelization codes
and the scrambling codes are expressed as
19
where DI(Q) denotes the I/Q components of the chip data sequence spread by
channelization codes and CI(Q) represents the I/Q components of a long
scrambling code. In this QPSK spreading, the carrier phase transition by
-degrees occurs across the zero point, thus incurring increasing nonlinear
distortion of the power amplifier. Therefore, in the 3GPP standard, the HPSK
(hybrid PSK) scheme was adopted, which decreased the possibility of the phase
transmission crossing the zero point [6, 51]. The long scrambling codes
sequence used for spreading are generated from the two original scrambling
codes based on the following equation:
In the forward link, P-SCH and S-SCH are spread by only primary and
secondary synchronization codes, respectively, commonly used for both I/Q
channels. The other physical channels except for SCH are first spread by an
identical channelization code with SF = m for both the I/Q channels and then
complex-scrambled by the two scrambling code sequences.
20
Improvements in W-CDMA
times the radio frame length (= 10 msec) and is defined as 10, 20, 40, or 80 msec
in the 3GPP. The transport format is a format in which a transport clock set is
transferred between the MAC layer and Layer 1 on a transport channel every TTI.
The transport format comprises two attributes: the dynamic part and semi-static
part. Attributes of the dynamic part are the transport block size, transport block
set size, and TTI, and those for the semi-static part are error of the correction
scheme such as the type of error correction and coding rate and the size of the
CRC. The transport format set (TFS hereafter) is defined as a set of transport
formats used for the transport channels. Within one TFS, the semi-static parts of
all transport channels are identical; however, the dynamic parts may be changed
every TTI in order to achieve variable rate transmission. The transport channels
are simultaneously multiplexed into Layer 1 as a coded composite transport
channel (CCTrCH). Each transport channel in the CCTrCH has an available TFS;
however, only one transport format is used at each TTI. Thus, the combination of
possible transport formats of all transport channels transferred on the same
Layer 1 at each TTI is defined as a transport format combination (TFC).
Furthermore, a set of TFC applied to the CCTrCH is called as transport format
combination set (TFCS). The indicator designating the TFC I called the transport
format combination indicator (TFCI). TFCI bits are multiplexed into the DPCCH of
each DPCH. In the receiver, the TFCI bits are used to decode Layer 1 data
sequences and de-multiplex transport blocks transferred on one physical
channel. In addition to the explicit TFCI detection method, the blind transport
format detection method using CRC to trace the surviving trellis path ending at
the zero state among the possible transport formats is also specified in the 3GPP
standard (note that blind detection is used only for the forward link) [9].
21
22
Improvements in W-CDMA
23
24
Improvements in
W-CDMA
where Ndatajis the total number of bits per radio frame to be assigned to code
the composite transport channel with TFC j and 1x1 denotes the integer value
defined as x - 1 s 1x1 5 x. Using the value of Zij recursively calculated from
Equation (6), ANij is derived from the following equation.
In the reverse link, rate matching is performed per radio frame based on
Equation (7). Meanwhile, in the forward link dissimilarly to the reverse link, DTX
is applied when there are no transmitted coded data bits of a certain transport
channel. Thus, the rate-matching pattern does not necessarily change for each
radio frame. Rate matching is performed as follows. The number of bits per TI
25
of transport channel i before rate matching, NTTIi,h, is first calculated for the
corresponding TFC h belonging to TFCS. Then, from the value of NTTIi,h, and th
number of radio frames of transport channel i over TTI, Fi, the corresponding
number of bits per radio frame was derived for all TFC belonging to TFCS. Thus
rate matching is performed such that the number of total bits per radio frame for
TFC hMax, when the summation of bits per radio frame of all transport channels
is maximized, is equal to the number of bits per radio frame accommodated into
a physical channel, that is to say, the number of bits per radio frame. Then the
number of bits per TTI to be bit-repeated or punctured is computed for each
transport channel. Based on this obtained rate matching pattern, the number of
bits per radio frame of each transport channel is updated every TTI.
Consequently, when transport channels having different TTI are multiplexed, the
number of total bits belonging to a radio frame is changed at the shortest TTI at
every TTI. If the number of bits per radio frame of transport channel i after rate
matching is lower than the maximum number of bits assigned to that transport
channel, DTX is performed during an interval corresponding to the number of bits
to be shortened.
26
Improvements in W-CDMA
length. The PSC for the primary-SCH is common to all cell sites and the ssc for
the secondary-sch denotes the group index into each of which the scrambling
codes are grouped beforehand. The total number of scrambling codes to be
searched is 512, which is divided into 64 groups of 8 codes each. The transmit
powers of the primary- and secondary-sch are set to half that of the CPICH.
The operational flow of the three-step cell search algorithm is illustrated in
Figure 8. Using SCHS and CPICH, the three-step cell search is performed as
follows. First, the PSC-matched filter (MF) is used. The MF output is averaged over
period T1 to detect the primary-sch time position that provides the maximum
average correlation. Next, the scrambling code group is identified by taking the
Figure 8.
Operational flow of
three-step cell
search method.
27
cross-correlation between the received signal and the set of sscs over period T2.
Finally, the scrambling code is identified by taking a partial correlation between
the received signal and each of the candidate scrambling codes and then
averaging over period T3. The scrambling code that provides the maximum
correlation is determined as the scrambling code to be searched. To reduce false
detection, a verification mode is added by using a frame synchronization check.
When the synchronization verification failed two consecutive times, the cell
search process is restarted from the first step. The correlation peaks of PSC and
ssc calculated in the first and second steps are averaged during T1 and T2 in
order to reduce the influence of MAI and the background noise components.
However, especially when the velocity of a MS is low, the probability for false
detection in the first and second steps is greater since the duration of low
received signal power due to fading becomes longer. Thus, time space transmit
diversity (TSTD) is applied to sc in the 3GPP specification, with which primaryand secondary-sc are alternatively transmitted slot-by-slot from different
antennas [5]. Since a successive primary- and secondary-SCH are transmitted
from different antennas having a low fading correlation, the false detection is
decreased due to the transmit diversity effect.
Figure 9 shows the measured laboratory experimental results of the
probability distribution of the cell search time with the fading maximum
Doppler frequency, fD, as a parameter using the 4.096 Mcps WCDMA
experimental system with TSTD [12, 13]. In addition to CPICH and schs, 10 DPCHS
without fast TPC were transmitted as a channel load. An L = 2 path Rayleigh
28
Improvements in W-CDMA
fading channel with average equal power was assumed because we confirmed
that field experimental results conducted near Tokyo could be well approximated
using this model where 2 - 3 paths with unequal average received signal power
were observed. The transmit power ratio of CPICH to DPCH and average received
Eb/No of DPCH were set to - 3 dB and 7 dB, respectively. We set T1, T2, and T3
to 40, 30, and 10 msec, respectively. Figure 9 shows that as fD becomes larger,
the cell search time becomes shorter since false detection is decreased. The
figure also shows that by using TSTD, the cell search time when fD is low such as
5 and 20 Hz can be decreased because false detection is mitigated when the
received signal level drops. As a result, the cell search time at the detection
probability of 90% with TSTD is decreased by approximately 100 msec compared
to that without TSTD. The cell search can be completed within approximately 250
msec at the probability of 90% with TSTD, when R = - 3 dB and fD = 5 Hz.
where
29
Then, Il(k) is averaged using a first order filter with forgetting factor (< 1 )
to obtain
The SIR at the k-th slot associated with the l-th path l(k) is given by
The measured SIR was compared to the target SIR and the TPC command w
generated, which was transmitted to raise or lower the mobile transmit power by
1 dB every 0.667 msec. Even if the received SIRS are the same, the received
quality (BLER) is not the same because the BLERS are affected by the number of
paths, maximum Doppler frequency (which depends on the speed of the
vehicle), and SIR measurement, etc. Therefore, the outer loop controls the targe
SIR with a more gradual updating interval compared to the inner loop so that the
measured BLER or BER is equal to the target value. In general, a BLER-based outer
30
Improvements in W-CDMA
loop is used. BLER is measured by calculating the number of CRC results that
coincide with the value attached to every transport block. Since the required
BLER becomes a very small value for high-speed and high-quality data
-6
transmission, e.g., with the required BER of 10 , it takes a much longer time to
calculate the BLER. As a result, outer loop control cannot track changes in the
propagation conditions. Therefore, in these cases, outer loop control based on
BER measurement of the tentative decision data symbols before channel
decoding (i.e., after Rake combining) with decision data symbols after channel
decoding as a reference can be applied. The reference data symbols are
generated by re-encoding and interleaving binary decision data symbols after
channel decoding. Although data decision error occurs in the decoded data
sequence, it is considered that the impact on the reference symbols is very small.
VI. Diversity
VI. 1. Coherent Rake combining
(Rake time diversity)
PSA coherent detection is used for both the reverse and forward links [16, 17].
The block diagram of the PSA coherent Rake combiner is illustrated in Figure
11 (a). The received multipath signals are despread by the MF and resolved into
L-multipath components of transmitted quaternary phase shift keyed (QPSK)
modulated data that are received via different propagation paths with different
delay times. The coherent Rake combiner output is expressed at the n-th symbol
position of the k-th slot associated with the l-th path (/ = 0,1, ..., L -1) using
despread signal r l (n, k), as
where l(k) represents the channel estimates. The output data sequence, d (n, k),
is de-interleaved and channel decoded to recover the transmitted binary data
sequence. In order to achieve accurate channel estimation that works
satisfactorily in a fast fading environment, we presented an improved channel
estimation filter called a weighted multislot averaging (WMSA) channel
estimation filter [17] as shown in Figure ll(b). After obtaining the instantaneous
channel estimates of each slot, the channel estimates, l(j + i)s, of 2J-multiple
slots (i = - J + 1,..., 0, 1, ..., J) are then weighted and summed to obtain the final
channel estimate, l(k), as
31
Receiv
spread
Figure 11. Coherent Rake receiver. ( a )Receiver structure (6) WWSA channel estirnationfrlter,
where ai is the real-valued weight. Using the WMSA channel estimation filter,
accurate channel estimation is possible, particularly in slow fading
environments. The optimum value of ai varies according to the fading
correlation between succeeding slots in a real fading channel. Therefore, we
proposed in [18] an adaptive WMSA channel estimation filter, in which a
weighting factor is adaptively controlled by measuring an inner product of the
averaged despread pilot signals of successive slots.
We evaluated the BER performance of coherent Rake combining with SIR
based fast TPC in field experiments conducted in an area near Tokyo. The cell site
and mobile transmitterh-eceiver antennas were located 59 and 2.9 m off the
ground, respectively. A measurement vehicle equipped with the mobile receiver
was driven along roads at distances of 0.75 - 1.35 km from the cell site at the
average speed of approximately 30 k d h . The measurement course passes
through a business zone, lined with office buildings and factories. Other
conditions are given in detail in [19]. The average delay spread of the test course
was approximately 1 psec. The test course first experienced clear two-path and
32
Improvements in W-CDMA
single-path fading at the middle of the course. Then, three-path fading with
unequal average power was observed at the end of the course.
Figure 12 plots the measured average BER performance of the 32-kbps data
rate user in the single-user and two-user cases (one-interfering user with a 64kbps data rate assuming the same BER independently employing fast TPC), as a
function of the TPC target Eb/I0 value (note Eb/I0 is calculated as Eb/I0 = SIR
101og(3/2) dB, since convolutional coding with the rate of 1/3 and QPSK data
modulation were used in the experiments) [19]. Two MSS established radi
links with BS 1. A WMSA channel estimation filter with J = 2 was used.
Laboratory experimental results of the single-user case using the L-path model
with fD = 80 Hz are also plotted for comparison. The results clearly show that
the target Eb/I0 when the interfering user exists becomes almost the same in
order to achieve the same BER as that of the single-user case, implying that fast
TPC worked satisfactorily in a real fading channel. The measured numbers of
active Rake fingers per antenna along test courses is 2.0. Figure 12 shows that
the measured BER performance is almost the same as the laboratory-measured
BER performance when L = 2. The field-measured BER performance results are
in good agreement with those estimated from the laboratory experiment. The
figure also shows that two-branch space diversity (antenna diversity) reception
can reduce the target Eb/I0 by approximately 3 dB at the average BER of 10-3.
33
With space diversity reception, the average BER of 10-3 can be achieved at the
required Eb/I0 of approximately 3 dB per antenna.
The measured average BER performance of the 64 kbps channel using turbo
coding is plotted in Figure 13 as a function of the MS relative transmit power
with the channel interleaving length of TCHL = 40 msec [20]. Turbo coding with
the rate of R = 1/3 and the constraint length of K = 4 bits (generator polynomials
are 13, 15, and 15 in octal notation) were used, while the rate and constraint of
convolutional coding as a reference were R = 1/3 and K = 9 bits, respectively.
Primary interleaving (PIL) [9, 22] and multistage interleaving (MIL) [9, 21],
which offer a greater capability for randomization compared to the block
interleaving method, were used as turbo interleaving and channel interleaving
methods, respectively. In the experimental system, Max-log - Map decoding
was used as the soft-in/soft-out decoder and the number of iterations, m, was
assumed to be eight, which was sufficiently large. From Figure 13, the MS
average transmit power for the average BER of 10-6 using turbo coding can be
decreased by approximately 0.6 (0.3) dB compared to that using convolutional
coding without (with) antenna diversity reception. Although the superiority of
turbo coding to convolutional coding was confirmed in an actual multipath
fading channel, this difference was decreased compared to the laboratory
experiments assuming a fixed delay time for each path using a fading simulator,
Figure 13. Average BER of 64-kbps data transmission with turbo coding
as a junction of mobile transmit power. Field experiments.
34
Improvements in W-CDMA
i.e., superiority was confirmed to be above 1.0 dB. This abatement in the
improvement with antenna diversity reception indicated that in an actual fading
channel in the field experiments, the impact of path search for Rake combing and
SIR measurement for fast TPC diminished the improvement in performance of the
turbo coding due to a very low received signal power level.
VI.2. Site diversity (soft/softer handover)
Soft handoff or site diversity ("site diversity" hereafter) [23, 24], which was first
implemented in the IS-95 CDMA standard [25], is an essential technique together
with fast TPC in improving transmission impairment due to multipath fadin
and shadowing near the cell edge. The simplified configuration of site diversity
is illustrated in Figure 14. In the forward link, the same original information
sequences before channel coding are transferred to NBSS(N is the number of BSs
with which the MS is associated) through the back haul (wired transmission lin
between BS and radio network controller (RNC)) from a RNC and transmittedfrom
two BSS using different scrambling codes. The received signals after Rake
combining at the MS are combined symbol-by-symbol with maximal ratio
combining (MRC) followed by soft-decision Viterbi decoding. With inter-sector
diversity in the reverse link, the Rake-combined signal of each sector is
combined with MRC in the same way as in the forward link. On the other hand
with inter-cell diversity, in the reverse link, a hard-decision data sequence after
soft-decision Viterbi decoding at each BS is transferred to the RNC via the back
haul with the reliability information associated with each traffic channel. The
transferred data sequences are selection-combined every selection period,
according to the reliability information.
35
36
Improvements in W-CDMA
Figure 15.
Instantaneous time variations in the reverse link inter-cell site diversity.
Field experiments.
37
transmit powers of the BSS causes a reduction in site diversity gain and an
increase in the interference to other users. To overcome this problem, several
schemes that compensate for the BS transmit power were proposed [28, 29]. I
the method proposed in [28], each BS controls its instantaneous transmit pow
by using a forgetting factor so that the difference between instantaneous transmit
power and the BS-specific reference transmit power calculated by averaging the
instantaneous values does not become large. However, it is difficult to quickly
track variations in path loss including shadowing due to the movement of the MS
The method in [29] reduces TPC bit error by sending the same TPC bit ov
several slots in the site diversity mode, this prevents the transmit power
difference between BSS from becoming large. However, in addition to the
problem described in [29], the TPC delay increases. Therefore, we proposed th
following two step algorithm to reduce the impact of TPC errors and keep the
transmit power of the BSS the same as that shown in Figure 16 [26].
(1) First loop: the standard transmit powers, P(k)REF, of all BSk are
compensated by P(k) according to the dedicated control channel from a MS
based on the average SIR measurement at a MS.
where the Measured_total_Eb/I0 and Target_ Eb/I0 are the measured Eb/I0 after
Rake combining and the target Eb/I0 at a MS, respectively. The P(k)REF is constan
during the length of G-slot and its value of n (= gxG)-th slot P(k)REF(n) is updated
every G-slot as P(k)REF(g x G) = P(k)REF((g-1) x G) +
38
Improvements in W-CDMA
(k)
The measured average BER performance in the forward link when inter-cell sit
diversity is applied is plotted in Figure 17 as a function of the total BS average
transmit power. The measurement course and experimental conditions were
the same as those in Figure 15. Fast TPC was used for the reverse and forwar
links. The number of maximum Rake fingers for the BS and MS was 4. The
forgetting factor was set to R = 0.8. In the measurement course, it was
observed that the instantaneous transmit power is controlled around the
standard transmit power without dispersing to the maximum output during the
course. The figure shows that the total transmit power of the 2 BSS at the
average BER of 10-3 in the inter-cell site diversity is decreased by
approximately 0.3 dB compared to a one cell-site connection. This
improvement is small compared to that in the reverse link because the increase
in interference due to transmissions from two BSS diminished the diversity
effect. Thus, the site selection diversity transmit power control (SSDT) [30], in
which only the primary BS transmits control bits to decrease the interference,
was proposed.
39
It is clear that the orthogonality between the two data sequences is maintained
irrespective of the spreading code sequence.
40
Improvements in W-CDMA
On the other hand, closed-loop type (two modes were standardized) transmit
diversity is used for DPCHS, in which the transmit antenna weights are controlled
by the FBI generated at the MS [33]. Let W1 = A1ej1 and W2 = A2ej2 be the
transmit antenna weights. Thus, in Mode 1, the transmitted phase of the second
antenna, 2, is changed with the accuracy of /4 according to the FBI from the M
so that the received SIR after combining is maximum. This is expressed as 1 = 0,
2 = {+/- /4, +/- 3/4}, A1 = A2 = v 1/2. Meanwhile, th transmitted
amplitudes of two data sequences are also controlled by FBI bits as well as th
transmitted carrier phase in Mode 2.
The measured average BER performance with STTD is plotted in Figure 1
when fast TPC was not applied in the forward link as a function of the averag
received Eb/I0, where I0 is the multipath interference plus background noise
power density [35]. The measurement course was course #1 described in [19].
The performance with and without antenna diversity reception at a MS is show
in the figures. The BER performance with signal antenna transmission is als
depicted for comparison. Figure 19 shows that the average required received
Eb/I0 at the average BER 10-3 with STTD was decreased by approximately 1.5
(1.0) dB without (with) antenna diversity reception. The improvement using
STTD with antenna diversity reception became smaller than that without antenna
diversity because the degradation of the channel estimation due to a lower
41
received level offset the additional diversity effect by STTD when using Rake
path diversity and antenna diversity reception. From the figure, the effectiveness
of STTD for a channel without TPC such as a common control channel was
elucidated in a real multipath-fading channel.
42
Improvements in W-CDMA
43
stage, are
weighted by the interference rejection weight control (IRWC) value, ap [47], and
removed from the signal received. It was confirmed by the some of the authors
[47] that by setting ap to less than one, the BER performance using COMSIC is
improved since the influence of the generation error of the interference replica
due to the channel estimation error and data decision error is mitigated. The
effect of IRWC is larger in the parallel-type COMSIC than in the serial-type. In our
COMSIC scheme, since the complex fading envelope (channel impulse response)
is recursively estimated and updated at each cancelling stage, the accuracy of
channel estimation and thereby that of the regenerated interference replica is
successively improved.
Figure 22 is a block diagram of the proposed CEIGU. In the proposed scheme,
the ICE using both pilot and decision-feedback data symbols after channel
decoding (hereafter forward error correction (FEC) decoding) [48] is applied to
each stage. Each CEIGU performs channel estimation, Rake combining, deinterleaving, Viterbi decoding, data decision, and replica generation of the MAI
for each user. Let kb,i(t), be the complex-valued channel impulse response of
the l-th path of the b-th antenna of the k-th user signal and(p,r)k,b,l(n)be its
estimated value after the r-th iteration (0 r R) in the p-th stage. In the first
stage, the initial channel estimate (estimated complex channel impulse response)
k,b,l(1,0)(n) of the n-th slot is computed by coherently averaging 2N pilot symbols
included in two successive slots. Then, a Rake combined data sequence is
obtained by coherently summing the Lk X B resolved signal components using
45
channel estimates k,b,l n ICE using both pilot and decision-feedback data
symbols after FEC decoding (hereafter ICE after FEC decoding), the Rake
(1,1)
combined sequence, d k (n, m), is de-interleaved and soft-decision Viterbi
(1,1)
decoded to recover the transmitted binary data sequence, b
(i) . The original
data sequence recovered is convolutionally-encoded and interleaved to generate
(1,1)
the binary decision-feedback data symbol sequence, d (n, m). Then, after the
data modulation of the MF output signal is removed by reverse-modulation usin
the decision-feedback data symbols, channel estimation using Nd decisionfeedback data symbols in addition to 2N pilot symbols is performed for each
slot. On the other hand, in ICE using both pilot and tentative-decision data
symbols before FEC decoding (hereafter ICE before FEC decoding), we use
binary tentative-decision data sequence after Rake combining (namely before
(1,1)
FEC decoding), sgn[d k (n, m)], instead of the decision-feedback data sequence
(1,R)
after FEC decoding in ICE after FEC decoding. L
n) and
(n, m) be t
channel estimate after the r-th iteration (1 r R), then the interference replica
of each path of each antenna is re-generated precisely as
where
Figure 22. Block diagram of CEIGU with ICE using pilot and decision-feedback data symbols.
47
The required average transmit Eb/N0 for achieving the average BER of 10 o
the parallel-type COMSIC using two-branch antenna diversity reception is plotted
in Figure 23 as a function of fD based on laboratory experiments using a fading
simulator. External Gaussian noise was added as background noise to the
receiver input. The number of active users was K = 3, and the SF of the desire
user and two-interfering users were SF = 64 and 8, respectively. The number o
stages was P = 3. The power delay profile used here was a two-path profile, each
path being subjected to independent Rayleigh fading with equal average power.
The target Eb/I0 ratio of the desired user to interfering user was set Eb/I0 = - 3
dB and - 6 dB. Thus, when Eb/I0 = - 6 dB, the received signal power of the
interfering user is 32-fold that of the desired user. The TPC delay was assumed t
be one slot. Figure 23 shows that the transmit Eb/N0 required using decision data
symbols after Rake combining with P = 3 is slightly decreased compared to that
using decision data symbols after channel decoding with P = 2 irrespective
of fD. The transmit Eb/N0 required using decision data symbols after Rake
combining with P = 2 is reduced by approximately 0.6 (0.8) and above 2.0 dB
compared to that of COMSIC using only pilot symbols and the MF-based Rake
receiver, respectively, when Eb/N0 = 3 (- 6) dB. Furthermore, the loss of th
required Eb/N0 from the single user case due to the channel estimation error and
Figure 23.
Average transmit Eb/N0 for the average BER 10-3 using parallel type comsic
as a function of fD (with antenna diversity reception, Eb = - 3, - 6 dB).
48
Improvements in W-CDMA
data decision error when Eb/N0 = - 3 dB was approximately 0.7 to 2.0 dB.
When the fD was small, the transmit Eb/N0 was slightly increased so that the fast
TPC should compensate for the drop in the signal level received due to fading.
Then, although the required transmit Eb/N0 was decreased as fD was increased
owing to the increasing interleaving effect associated with channel coding, it was
significantly degraded due to the degradation in the channel estimation tracking
ability against fast fading when fD was larger than approximately 150 Hz.
VII.2. AAAD
VII.2.1. CAAAD receiver/AAA-TD
We also investigated the interference suppression effect in the reverse and
forward links using the experimental CAAAD receiver/AAA-TD [40-45]. The
carrier frequencies were the same as those of the COMSIC. The spreading chip rate
was 4.096 Mcps and the bandwidth was 5 MHz. Figure 24 shows an overall
block diagram of the four-antenna AAAD transceiver. The receiver antenna
weights in the reverse link are generated in the CAAAD receiver block. In our
scheme, transmitter antenna weights are generated by modifying the receiver
antenna weights (i.e., performing calibrations we will be explaining later)
[40-43]. Our concern is frequency division duplex (FDD) in W-CDMA systems,
thus, the instantaneous phase and amplitude variations due to fading in the
reverse link have no correlation with those in the forward link. Noting that the
distance-dependent macroscopic propagation factors determining the average
signal power, i.e., path loss and shadowing, are not frequency dependent, we
designed the CAAAD receiver based on the following concept. The adaptive
antenna array forms an antenna beam that tracks only slow changes in the
directions of arrival (DOAs) and average powers of the desired and interfering
users, and the Rake combiner tracks the instantaneous variations in the channel
conditions to maximize instantaneous signal-to-interference plus background
noise power ratio (SINR).
We perform two calibrations to the receiver antenna weights in generating
transmit antenna weights in the forward link. We use the common array antennas
with the antenna separation of half the carrier wavelength in the reverse link.
Thus, when the DOA of the incoming signal is not 0 degrees, the phase difference
between antennas is different between different carrier frequencies. If the
generated receiver antenna weights are used directly as transmitter antenna
weights, the direction of the main lobe in the transmitter beam pattern is shifted
from the original DOA of the desired user and the directions of the beam nulls are
also shifted from those of interfering users. Therefore, in our approach, we shift
the direction of the main lobe in the transmitter beam pattern so that it coincides
50
Improvements in W-CDMA
with the main lobe in the receiver beam pattern. We call this modification in the
transmitter beam pattern carrier frequency calibration.
On the other hand, since adaptive antenna array processing is done in the
baseband, the generated reverse link weights are reflected on the
phase/amplitude deviations in the transfer functions of the radio frequency (RF)
receiver circuitry associated with different antennas. In the transmitter, the RF
transmitter circuitry of the different antenna branches is adversely influenced by
the different transfer functions (see Figure 24). Therefore, the transmit weights
are generated by modifying the receiver antenna weights taking into account the
phase/amplitude variations in RF receiver/transmitter circuitry. This calibration i
called RF circuitry calibration. Let w(i)ideal, and w(i)R andx(i)RXbe the complex-value
antenna weight for the ideal case (no phase/amplitude errors exist in the RF
receive circuitry transfer functions), the complex-valued receiver antenna
weight generated in CAAAD, and complex-valued transfer function of the RF
receive circuitry of the i-th antenna branch, respectively. For the ideal case (i.e.,
x(i)RX=const.for all antenna branches), received signal r(i)RX should be weighted by
w(i)ideai to be combined for beam forming. However, in a real receiver, r(i)RX goes
through the RF receiver circuitry and then, weighted by w(i)R, produces w(i)R x(i)R
r(i)RX, before combining. The resultant signal must be the same as that of th
ideal case. Therefore, we obtain
The weights provided by Equation (23) cannot be directly used to form the
transmit beam because the signal weighted at the baseband stage suffers
phase/amplitude shift due to the RF transmit circuitry before transmission from
an antenna. Let x(i)TX be the transfer function of the RF transmit circuitry for the
i-th antenna branch. The signal to be transmitted from the i-th antenna is first
weighted using transmit antenna weight w(i)T and then suffers phase/amplitude
shift (equivalent by multiplication of x(i)TX). The equivalent antenna weigh
becomes, therefore, w(i)T . x(i)TX, and this must be equal to w(i)idea/. As a consequenc
we obtain
Using Equation (24), the transmit antenna weights (at the base band beam
forming stage) can be obtained from adaptively generated receive antenna
weights. Bothx(i)RXandx(i)TXcan be measured even during the operation mode. By
performing the above two calibrations on the generated receiver antenna weights
in the CAAAD receiver, the maximum gain is obtained toward the desired signal
direction and the nulls are nearly directed towards the interfering sources.
51
Figure 25 shows a block diagram of the digital beam former and Rake
combiner in the CAAAD receiver. This block comprises MFS, a beam former, PSAcoherent Rake combiner, and weight controller. The MF output signal samples of
all antennas for each resolved path are weighted by the receiver antenna weight,
which is common to all paths for Rake combining, and then combined. Since it
was reported that the angle spread among paths is within ten degrees in an urban
area with a high elevation antenna configuration such as in a cellular system [49,
50], we used the common antenna weights, i.e., common receiver beam pattern,
for all paths. The resulting composite channel gain at the beam former output is
estimated by using the pilot symbols multiplexed into the Q-channel of three
successive slots for coherent Rake combining of different resolved paths. The
receiver antenna weights in the CAAAD receiver are updated so that the mean
squared error (MSE) between Rake-combined signal and the reference signal is
minimized. As a reference signal, we used the decision-feedback data symbols
after FEC decoding in addition to pilot symbols for generating refined MSE an
improving channel estimation accuracy.
VII.2.2. Experiments
First, we investigated the effect of receiver antenna weight updating using
decision-feedback data symbols after FEC decoding in addition to pilot symbols
and performance difference for channel coding. The average BER performance
measured with the CAAAD receiver with fast TPC using convolutional coding and
turbo coding is shown in Figure 26 as a function of the average transmit E b /N 0
in the reverse link. The DOAS of the desired and interfering users were set to 0
and + 40 degrees, respectively, in a 2-user environment. The ratio of the target
Eb/I0 for fast TPC of the desired user to interfering user was E b /I 0 = - 12 dB
The performance with receiver weight updating using decision-feedback data
symbols as well as pilot symbols, and pilot symbols only are plotted as solid and
dotted lines, respectively. It was assumed that fD is 5 Hz and TCHL is 20 msec.
Figure 26 shows that the transmit Eb/N0 required for satisfying the average BER
of 10-3 and 10-6 with the CAAAD receiver using decision feedback antenna
weight updating and channel estimation is decreased by approximately 0.8 dB
compared with the case using only pilot symbols both for convolutional coding
and turbo coding. These results also confirm that for TCHL= 20 msec, the required
transmit Eb/N0 at the average BER of 10-6 of the CAAAD receiver using decision
feedback antenna weight-updating and channel estimation with turbo coding is
decreased by approximately 0.5 dB compared to the case with convolutional
coding. Thus, the results given here agree well with those from the MF-based
Rake receiver in Figure 13.
53
Figure 26.
Average BER performance as a function of average transmit Eb/N0
(with fast TPC, 2 users).
54
Improvements in W-CDMA
55
Figure 28. Average BER performance as a function of average received signal power.
Field experiment.
56
Improvements in W-CDMA
VIII. Conclusion
This paper reviewed the channel structure and spreading code assignment in
the physical layer, transport channel multiplexing into a physical channel
along with sophisticated rate matching, and the key technologies of W-CDMA
wireless access. Experimental results of the three-step cell search method
associated with TSTD exhibited fast cell search time performance, while
flexible continuous system deployment from outdoors to indoors was possible.
Pilot symbol based coherent Rake receiver exhibited efficient, i.e., lower
required Eb/I0, transmission together with fast TPC and space diversity
reception based on experiments in actual multipath fading channels. We also
verified the effectiveness of site diversity and transmit diversity in decreasing
the transmit power, thereby increasing link capacity. These experimental
results associated with the successful system level experiments showed the
superiority of W-CDMA wireless access technologies for commercial services.
Link capacity enhancing techniques such as those employing a COMSIC and
CAAAD receiver/AAA-TD and examples of BS transceiver using these techniques
were also presented with experimental results. The laboratory and field
experimental results clarified the potential of the COMSIC and CAAAD receiver to
reduce the mobile transmit power in the reverse link and that of AAA-TD to
decrease MPI with large transmit power in the forward link. These techniques ar
essential for achieving high-speed packet transmission based on the W-CDMA air
interface in an actual severe multipath fading channel.
Acknowledgements
The authors thank K. Okawa, S. Fukumoto, A. Morimoto, H. Taoka, and A.
Harada of the Wireless Access Laboratory for their contributions to the W-CDMA
experiments.
57
REFERENCES
[1] ADACHI (F.), SAWAHASHI (M.), and SUDA (H.), "Wideband DS-CDMA for Next
Generation Mobile Communication System," IEEE Commun. Mag., Vol. 36,
pp. 56-69, (Sept. 1998).
[2] DAHLMAN (E.), GUDMUNDSON (B.), NILSSON (M.), SKOLD (J.), "UMTS/IMT-2000 Based on
Wideband CDMA," IEEE Commun. Mag., Vol. 36, pp. 70-80, (Sept. 1998).
[3] SAWAHASHI (M.), HIGUCHI (K.), TANAKA (S.), and ADACHI (F.), "Enhanced Wireless
Access Technologies and Experiments for W-CDMA Communications,"
IEEE Personal Communications, vol. 7, n 6, pp. 6-17, (Dec. 2000).
[4] 3GPP RAN, 3G TS 25.301 V3.5.0, (June 2000).
[5] 3GPP RAN, 3G TS
[6] 3GPP RAN, 3G TS
[7] OKAWA (K.) and
Multi-spreading
58
Improvements in W-CDMA
[18] ABETA (S.), SAWAHASHI (M.), and ADACHI (F.), "Adaptive Channel Estimation for
Coherent DSCDMA Mobile Radio Using Time-Multiplexed Pilot and Parallel Pilot
Structure" IEICE Trans. Commun., Vol. E82-B n 9, pp. 1505-1513, (Sept. 1999).
[19] HIGUCHI (K.), ANDOH (H.), OKAWA (K.), SAWAHASHI (M.), and ADACHI (F.),
"Experimental Evaluation of Combined Effect of Coherent Rake Combining and
SIR-based Fast Transmit Power Control for Reverse Link of DS-CDMA Mobile Radio
IEEE J.Select. Areas Commun., Vo.18, pp.1526-1535, (Aug. 2000).
[20] HIGUCHI (K.), IKEDA (T.), FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.),
"Experimental Evaluations of High Rate Data Transmission Using
Turbo/Convolutional Coding in W-CDMA Mobile Radio" Proc. Wireless 2000, Calgary,
pp. 687-693, (July 2000).
[21] SHIBUTANI (A.), SUDA (H.), and ADACHI (F.), "Multi-stage Interleaver for Turbo Codes
in DS-CDMA Mobile Radio," Proc. IEEE APCClICC'98, pp. 391-395, (Nov. 1998).
[22] SHIBUTANI (A.), SUDA (H.), and YAMAO (Y.), "Performance of
W-CDMA Mobile Radio with Turbo Codes Using Prime Interleaver," Proc. VTC '2000,
Tokyo, Japan, pp. 1570-1574, (May 2000).
[23] GILHOUSEN (K.S), JACOBS (I.M), PADOVANI (R.), VITERBI (A.J), WEAVER (L.A), and
WHEATLEY III (C.E.), "On the Capacity of a Cellular CDMA System," IEEE Trans. Veh.
Technol., Vol. VT-40, n 5, pp. 303-312, (May 1991).
[24 WONG (D.) and LIM (T.J), "Soft Handoffs in CDMA Mobile Systems," IEEE Personal
Commun. Mag., Vol. 36, pp. 6-17, (Dec. 1997).
[25] TIA/EIA/IS-95, "Mobile Station-Base Station Compatibility Standard for Dual-Mode
Wideband Spread Spectrum Cellular System," Telecommunication Industry
Association, (July 1993).
[26] FUKUMOTO (S.), HIGUCHI (K.), MORIMOTO (A.), SAWAHASHI (M.), and ADACHI (F.),
"Combined Effect of Site Diversity and Fast Transmit Power Control in W-CDMA
Mobile Radio," Proc. VTC '2000, Tokyo, Japan, pp. 1527-1534, (May 2000).
[27] MORIMOTO (A.), HIGUCHI (K.), FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.),
"Experiments on Inter-Cell Site Diversity Using Two-step Selection Combining in
W-CDMA Reverse Link," IEICE Trans. Commun., Vol. E84-B n 3, pp. 435-445,
(March 2001).
[28] NEC, "Adjustment Loop in Down Link Power Control During Handover," TSG-RAN
Working 1 Meeting, (Oct. 1999).
[29] Nortel, "Down Link Power Control During Soft Handover," 3GPP TSG-RAN WGI
Meeting, (July 1999).
[30] FURUKAWA (H.), HAMABE (K.), and USHIROKAWA (A.), "SSDT- Site Selection Diversity
Transmission Power Control for CDMA Forward Link," IEEE J.Select.Areas Commun.,
Vol. 18, pp. 1546-1554, (Aug. 2000).
[31] HOTTINEN (A.) and WICHMAN (R.), "Transmit Diversity by Antenna Selection in CDMA
Downlink," Proc. IEEE issta'98, pp. 767-770, (Sept. 1998).
[32] FUKUMOTO (S.), SAWAHASHI (M.), and ADACHI (F.), "Performance Comparison of
Forward Link Transmit Diversity Techniques for W-CDMA Mobile Radio,"
Proc. PlMRC'99, pp. 1139-1143, (Sept. 1998).
[33] 3GPP RAN, 3G TS 25.214 V3.4.0, (Sept. 2000).
59
[34] ALAMOUTI (S.M), "A Simple Transmit Diversity Technique for Wireless
Communications," IEEE J. Sel. Areas Commun., Vol. 16, n 8, pp. 1451-1458,
(Oct. 1998).
[35] FUKUMOTO (S.), HIGUCHI (K.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on Space
Time Block Coding Transmit Diversity (STTD) in W-CDMA Forward Link," Submitted
to IEEICE Trans. Fundamentals. Special Issue on Spread Spectrum.
[36] DUEL-HALLEN (A.), HOLTZMAN (J.), and ZVONAR (Z.), "Multiuser Detection for CDMA
Systems," IEEE Personal Communications, pp. 46-58, (April 1995).
[37] MOSHAVI (S.), "Multi-user Detection for DS-CDMA Communications," IEEE Commun.
Mag., pp. 124-136, (Oct. 1996).
[38] TSOULOS (G.V.), BEACH (M.A.), and MCGEEHAN (J.), "Wireless personal
communications for the 21st century: European technological advances in
adaptive antennas," IEEE Commun. Mag., pp. 102-109, (Sept. 1997).
[39] COMPTON (R.T.), Jr., "An Adaptive Antenna in a Spread-spectrum Communication
System," Proc. IEEE, Vol. 66, n 3, pp. 289-295, (March 1978).
[40] HARADA (A.), TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Performance of
Adaptive Antenna Array Diversity Transmitter for W-CDMA Forward Link,"
IEEE PIMRC'99, pp. 1134-1138, Osaka, (Sept. 12-15 1999).
[41] TANAKA (S.), HARADA (A.), SAWAHASHI (M.), and ADACHI (F.), "Transmit Diversity Based
on Adaptive Antenna Array for W-CDMA Forward Link," Proc. CDMA International
Conference (CIC)'99, pp. 282-286, Korea, (Sept. 1999).
[42] HARADA (A.), TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on
Adaptive Antenna Array Transmit Diversity in W-CDMA Forward Link," Proc. CDMA
International Conference (CIC)'2000, pp. 47-51, Korea, (Nov. 2000).
[43] TAOKA (H.), TANAKA (S.), IHARA (T.), and SAWAHASHI (M.), "Experiments on Adaptive
Antenna Array Transmit Diversity with Carrier Frequency Calibration in Transmit
Power-Controlled Forward Link for W-CDMA Mobile Radio," Proc. 3G Wireless and
Beyond, San Francisco, (May 2001).
[44] TANAKA (S.), SAWAHASHI (M.), and ADACHI (F.), "Pilot Symbol-assisted Decisiondirected Coherent Adaptive Array Diversity for DS-CDMA Mobile Radio Reverse
Link,"IEICETrans. Fundamentals, Vol. E80-A, pp. 2445-2454, (Dec. 1997).
[45] TANAKA (S.), HARADA (A.), SAWAHASHI (M.), and ADACHI (F.), "Experiments on
Coherent Adaptive Antenna Array Diversity for Wideband DS-CDMA Mobile
Radio," IEEE J.Select.Areas Commun., Vol. 18, pp. 1495-1504, (Aug. 2000).
[46] SAWAHASHI (M.), MIKI (Y.), ANDO (H.), and HIGUCHI (K.), "Pilot Symbol-Assisted
Coherent Multistage Interference Canceller Using Recursive Channel Estimation
for DS-CDMA Mobile Radio," IEEICE Trans. Commun., Vol. E79-B, pp. 1262-1270,
(Sept. 1996).
[47] SAWAHASHI (M.), ANDOH (H.), and HIGUCHI (K.), "Interference Replica Weight
Control for Pilot Symbol-Assisted Coherent Multistage Interference Canceller
Using Recursive Channel Estimation in DS-CDMA Mobile Radio," IEICE Trans.
Fundamentals, Vol. E81-A, pp. 957-972, (May 1998).
60
Improvements in W-CDMA
[48] OKAWA (K.), HIGUCHI (K.), and SAWAHASHI (M.), "Parallel-type Coherent Multi-stage
Interference Canceller with Iterative Channel Estimation Using Both Pilot and
Decision-feedback Data Symbols for W-CDMA Mobile Radio," IEICE Trans.
Commun., Vol. E84-B n 3, pp. 446-456, (March 2001).
[49] KOZONO (S.) and SAKAGAMI (S.), "Correlation coefficient on base station diversity
for land mobile communication system," IEICE Trans Commun., Vol. J70-B,
pp. 476-482, (in Japanese), (Apr. 1987).
[50] MARTIN (U.), "Spatio-temporal radio channel characteristics in urban macrocells,"
IEEE Proc.- Radar, Sonar Navig., Vol. 145, (Feb. 1998).
[51] IMT-2000 Study Committee, AIF/SWG2-25-4, (July 1998).
Chapter 3
I. Introduction
Future multimedia services (Internet, video transmission, data transfer, ...) will
require the transmission of very high data rates over broadband radio channels.
In order to meet the quality of service (QOS) requirements of these future
multimedia applications, the multiple access techniques of next generation
cellular systems should offer flexibility and especially very high spectral
efficiency.
Spread spectrum has been very successfully used by the military services for
decades and more recently has taken on a significant role in cellular and personal
communications [1, 2]. Advantages of spread spectrum techniques are widely
known: immunity against multipath distortion and jamming, low transmitted
power, no need for frequency planning and last but not least, Code-Division
Multiple Access (CDMA) capabilities [3]. As a result of the discussion in the last
decade, CDMA will play an important role in the next generation of cellular
systems, i.e. in UMTS and in IMT-2000 [4]. Thus, Wideband CDMA is a leading
candidate for third generation wireless access. Based on direct sequence spread
spectrum with a chip rate of 3.84 Mchip/sec, the transmitted signal occupies a
bandwidth of about 5 MHz. It supports circuit and packet data access at nominal
peak data rate equal to 384 Kbit/sec for macro cellular environments and to
2 Mbit/sec for indoor environments. However, we already know that wireless
Internet usage is likely to be downlink limited. Indeed, for data services, peak bit
62
63
MC-CDMA
64
Figure 1.
MC-CDMA transmitter scheme and power spectrum of the transmitted signal.
effect of spreading is that different users can have access to the same carriers in
a CDMA manner. The separation of the user's signals is then performed in the
code domain. The advantage of MC-CDMA in comparison with DS-CDMA is that the
spreading can be adapted to the frequency selective behaviour of the channel.
Simple methods for signal detection in the frequency domain as one-tap
equalizer per carrier can be used.
Figure 1 shows the MC-CDMA transmitter of the jth user and the power
spectrum of the transmitted signal.
The data symbol xj. (t) of the user j is transmitted in parallel over Nc
subcarriers, each multiplied by one chip ck, j of the spreading code Cj (t) = [c1,j.,
c 2,j , ..., cLmc, j] assigned to user j. In this figure, the length Lmc of the spreading
code is equal to the number NC of subcarriers but this is not mandatory. As a
consequence, the MC-CDMA systems offer an additional degree of freedom, and
actually the number Nc of subcarriers is chosen to guarantee frequency nonselective fading over each subcarrier.
65
The expression of the transmitted signal Sj (t) of user j during the time
interval [0, TX] is:
where Tx is the data symbol duration which is in this case equal to the OFDM
symbol duration, xj the data symbol transmitted during the signalling interval
[0, Tx], P (t) the pulse shaping waveform which is generally rectangular and f0
is the carrier frequency.
Practically, the MultiCarrier modulation and demodulation is easily carried
out in the digital domain by performing IFFT and FFT operations. Furthermore the
insertion between adjacent MultiCarrier (MC) symbols of a guard interval A,
longer than the delay spread of the impulse response of the channel, guarantees
the absence of Inter Symbol Interference (ISI). In this case, the resulting
spectrum of the transmitted signal satisfies the orthogonality condition with the
minimum frequency separation equal to 1/TX = 1/(T' X -), where T'x is the new
MC symbol duration and Tx is the useful duration of this symbol. In the receiver,
after direct FFT and possibly de-interleaving, the received sequence is
"equalized" in the frequency domain. Therefore, the MC-CDMA receiver can
always employ all the received signal energy spread in the frequency domain.
Undoubtedly, this is the main advantage of the MC-CDMA scheme compared to a
DS-CDMA Rake receiver that has difficulties in making full use of the received
signal energy scattered in the time domain.
For a synchronous system as the downlink mobile radio communication
channel, the application of orthogonal codes such as Walsh-Hadamard codes
guarantees the absence of Multiple Access Interference (MAI) in a gaussian
channel. However, in non-ideal channels with frequency selective fading due to
multipath propagation, the orthogonality between the signals of the different
users is lost and MAI occurs. To combat the channel fading and thus the MAI, a
multitude of detection techniques was proposed. They can be classified as either
single-user detection (SD) or multi-user detection (MD). Single-user detection is
performed by one tap equalization to compensate for the phase and amplitude
distortions caused by the mobile radio channel. The one tap equalizer is simply
one complex-valued multiplication per subcarrier. Basic SD techniques are
Maximum Ratio Combining (MRC), Equal Gain Combining (EGC), Orthogonal
Restoring Combining (ORC) or Minimum Mean Square Error (MMSE)
equalization. This last technique, based on the MMSE criterion applied
independently to each subcarrier achieves better performance [21, 25] as we
66
will see in section III. Another way to mitigate the MAI is to optimize the
spreading sequence selection within a given spreading sequence family as
presented in [26].
With the aim to improve the performance of the receiver still further, Multiuser Detection (MD) can be processed, where the a priori knowledge about the
spreading codes of the interfering users is exploited in the detection process.
Indeed, the receiver jointly detects the other active user signals in order to
mitigate the MAI. Multi-user detection methods can be divided into Interference
Cancellation (IC) and Joint Detection (JD). The principle of IC is to detect the
information of the interfering users and to reconstruct the interfering
contribution in order to subtract it from the received signal.IC can be performed
parallel for all interfering users with Parallel Interference Cancellation (PIC)
detectors, or successively with Successive Interference Cancellation (SIC)
detectors where only the strongest interferer remaining after the previous IC
stage is cancelled.
The optimal detector applies Joint Detection (JD) with Maximum Likelihood
Sequence Estimation (MLSE) or Maximum Likelihood Symbol-by-Symbol
Estimation (MLSSE), respectively. Since the complexity of MLSE and MLSSE
receivers grows exponentially with the number of users, their use is limited in
practice to applications with a small number of users.
Most MC-CDMA systems were considered for optimum use of the available
bandwidth of the synchronous downlink of a cellular network using orthogonal
codes as Walsh-Hadamard codes. For the uplink, another scheme called Spread
Spectrum Multi-Carrier Multiple Access (SS-MC-MA), which is derived from the
MC-CDMA concept, has been proposed in [27]. In this case, each user k,
exclusively transmits on a set of L subcarriers out of a total of NC carriers.
II.2. MC-DS-CDMA
67
Figure 2.
MC-DS-CDMA transmitter scheme and power spectrumof the transmitted signal.
If the spreading code length Lmd is inferior or equal to the number of subcarriers
Nc, a single data symbol is spread in the time domain. By using a high number
of subchannels, this concept benefits from time diversity. However, due to the
frequency non-selective fading per subchannel, frequency diversity can only be
exploited if channel coding with appropriate interleaving or subcarrier hopping
is used, or if the same information is transmitted in parallel on several
subcarriers [29]. In [30], a MC-DS-CDMA scheme with a subcarrier spacing larger
68
than the inverse of the chip duration is proposed in order to increase the
frequency diversity of the system.
In reception, the MC-DS-CDMA system is composed of Nc coherent receivers
for data detection.
II.3. MT-CDMA
with
which is identical to the expression of the MC-DS-CDMA transmitted signal given
by equation (2) except that the subcarrier separation is equal to l/(N c TX) instead
of Lmd/(Nc Tx). The MT-CDMA receiver is composed of Nc Rake combiners, each
of which has the same structure as the DS-CDMA Rake receiver. The MT-CDMA
scheme suffers from inter-subcarrier interference, while the capability to use
longer spreading codes, compared to the spreading codes assigned to a CDMA
scheme, results in the reduction of Self Interference (SI) and MAI. In a channel
where this improvement is dominant, the MT-CDMA scheme can outperform the
DS-CDMA scheme [17]. This concept was mainly investigated for an asynchronous
up-link scenario.
Advances in
MT-CDMA transmitter
UMTS
technology
Figure 3.
scheme and power spectrum of the transmitted signal.
I
I
Svstem
Spreading
MC-CDMA
MT-CDMA
Time domain
Time domain
Spreading length
Lm
Lmd-Lm
Symbol duration
at subcarrier
Tx
N, . T, I L-
N, . Tx
Subcarrier
separation
1 I TI
IlN,.T,
TI
( N , TJ J L ,
( N , t l)L,/
0 3 2 Lmr+N-1J
Chip duration
I MC-DS-CDMA I
I Frequency domain I
Required
bandwidth
(main lobe)
Reception
techniques
Main
applications
I
1
( N , + 1 ) T,
Downlink
-Quasi- synchm
nous uplink
I
I
( N c . TI)
- Coherent
Asynchronous
-uplink
4NJJ
Roll-off factor = 0
- Rake combiner
-MlMOequalizer
-Asynchronous
plink
69
70
Since 1993, especially the first and the second schemes have been studied
deeply. The capacity of the three systems is equivalent if Lmc Lmd (Lmt/Nc)
Nc. When a Nyquist filter with a small roll-off factor is used in the MT-CDMA
scheme, the MC-CDMA, MC-DS-CDMA and MT-CDMA systems require almost the
same frequency bandwidth to transmit the same bit rate 1/Tx. Deep system
analysis and comparison with DS-CDMA have been performed showing the
superiority of MC-CDMA [17-19]. In particular, for the synchronous downlink, the
MC-CDMA system with MMSE detection outperforms all other schemes and offers
the best trade-off between the spectrum and the power efficiencies.
As pointed out in the introduction, wireless Internet access is expected to grow
rapidly in the near future. This means that spectrum efficiency of the downlink
becomes crucial for these high-speed data services. In the following sections, the
performances of MC-CDMA systems are therefore considered for the downlink of
the future wideband wireless networks.
71
where the Nc x Nc diagonal matrix H = diag{h 1 , ..., hNc} describes the complex
channel frequency response and N = [n1, n2, ..., nNc]T is the vector containing
the Additive White Gaussian Noise (AWGN) terms with nk representing the noise
term at the subcarrier k with variance given by
=N2=E{\nk\2},k = 1, ..., Nc.
After equalization, the received signal can be written as:
72
where X represents the useful signal part, the MAI and the noise term.
73
Since the codes are orthogonal, half of the products ck, ick, j, i j, are equal to
1/Lmc
and the other half are equal to -\/Lmc
. Then, the variance of the multiple
mc
mc
access interference term can be written as:
and the variance of the additive zero mean Gaussian noise is given by:
In the following, the values of E, 2 and 2 are presented for various single
user detection techniques when the fading of the channel is Rayleigh distributed.
Maximum Ratio Combining technique (MRC)
In the single user case, MRC is the optimum diversity combining technique. The
corresponding equalization coefficients are:
74
The variances of the multiple access interference term and the additive zero
mean Gaussian noise are respectively:
75
with
As already reported in Section III. 1, the matrix G is diagonal for all these
basic single user detection techniques which means that the sequence received
is equalized by using a bank of Nc adaptive one tap equalizers which results in a
low complexity equalizer.
III.2.2. Analytical and simulation results with single user detection
techniques
The analytical and simulation results with MRC, EGC and MMSE detections are
presented in Figure 5 for a Rayleigh channel. The number Nu of active users is
equal to the spreading code length Lmc (full user capacity) which is also equal to
the number Nc= 64 of subcarriers.
The Matched Filter (MF) bound or "limit bound" is given as reference. It
corresponds to the BER obtained in the case of MRC detection for a MC-CDMA
system with only one user and consequently not affected by MAI.
It can be seen that the analytical results are really close to the simulation
results, particularly for MRC and EGC detections. For MMSE detection, there is a
76
77
describe the two types of IC detectors based on either a serial (SIC) or a parallel
(PIC) interference cancellation [32, 33].
The aim of each of these IC detectors is to estimate the contribution of
each user in the MAI term and to suppress it. The process may be performed
iteratively in multiple detection stages. The useful received signal is first
equalized by a SU technique and then despread by each code. Since the
downlink is considered synchronous, an Inverse Fast Hadamard Transform
(IFHT) may be carried out to reduce the receiver complexity.
IV.1. Successive Interference Cancellation
The SIC detector first detects the most powerful interfering user and then cancels
its contribution from the received signal. The second strongest interferer is then
cancelled and so on. The processing may be repeated for a few or for all users.
A complete detector would consider all users, but only the interferers stronger
than the useful one are commonly suppressed, SIC detector is generally used
when the power of some users are higher than the power of the useful user. Since
processing one supplementary stage leads to an additive time delay, a trade-off
between the number of stages and the total acceptable delay has to be found.
Figure 6 shows the scheme of one stage SIC detector where the strongest
interfering user is detected among the estimations x1 ... XNU and then cancelled
from the received signal. The process is carried out iteratively until the remained
interferers are considered insignificant. The resulting signal is finally despread.
The data detection may be hard or soft.
78
time delay of a SIC detector. The expression of this iterative system for the
m-th stage and the j-th user is given by the following:
79
NC
where R, R is the autocorrelation matrix of the received vector R and R,xj is the
cross-correlation vector between the desired symbol x. and the received signal
vector R. These terms are equal to:
The subcarrier noises have the same variance and are independent. Thus,
E{NN*T} = 2N,.1, where / is the identity matrix. Since the user signals have the
same power (E{x j 2 } = ES) and are independent, we can write E{XX*T} = Es.A,
where A = {aij} is a diagonal matrix with the term ajj = 1 if the user j is active
and ajj = 0 if the user j is inactive. Then, the equalization coefficient matrix is:
In the full load case (Nu = Lmc) and only in that case, the quantity C.A.CT is
equal to the identity matrix and the equalization coefficients matrix G is a
diagonal matrix with the k-th subcarrier equalization coefficient equal to the
former Equation (19).
80
On the other hand, when the capacity is not full (Nu < Lmc), the equalization
coefficient matrix G is no more diagonal. In that case, as will see in the last
section, the Global MMSE (GMMSE) algorithm outperforms the MMSE per carrier
algorithm, since it minimizes the decision error taking into account the
de-spreading process instead of minimizing the error independently on each
subcarrier.
Vehicular A
Vehicular B
Delay sec Power dB Delay sec Power dB Delay sec Power dB Delay sec Power dB
0.0
-7
0.0
-10
0.0
0.0
0.2
-3
0.2
-8
0.31
-1
0.3
0.0
0.4
-1
0.4
-6
0.71
-9
8.9
-12.8
0.8
0.8
-4
1.09
-10
12.9
-10.0
1.6
-2
1.6
1.73
-15
17.1
-25.2
2.2
-6
2.0
2.51
-20
20.0
-16.0
3.2
-7
2.4
-4
5.0
-1
15.0
-8
6.0
-2
15.2
-9
7.2
-7
15.8
-10
8.2
-10
17.2
-12
10.0
-15
20.0
-14
-2.5
81
HT
VA
VB
Channel
BU
FO
1.8 GHz
1.8 GHz
2 GHz
2 GHz
1 m/sec
27.7 m/sec
27.7 m/sec
27.7 m/sec
6 Hz
166.7 Hz
185 Hz
185 Hz
Tm
10. sec
20. sec
2.5 sec
20. sec
20. sec
20. sec
5. sec
20. sec
Tx
128. sec
128. sec
33.33 sec
133.33 sec
256
256
256
1024
11
0.793 bit/sec/Hz
0.793 bit/sec/Hz
0.797 bit/sec/Hz
0.797 bit/sec/Hz
2 MHz
2 MHz
7.68 MHz
7.68 MHz
Dmax
Nc
The last parameters of the table depending on the system are given for each
channel:
-
82
Furthermore, all users are considered to have the same power, which means
that power control may be implemented. For the comparison of the different
detection systems, Bit Error Rate (BER) performance is provided as a function of
Eb/N0 where Eb is the energy per useful bit. All results have been obtained
through simulation with the Monte Carlo method. For most results,
2 limit curves are provided:
a lower limit called "limit limit bound" that stands for a system with only
one user and a matched filter (MRC detection) and consequently without
MAI,
a upper limit that corresponds to a "full load" system with the maximum
MAI.
83
84
The total bit rate increases with the number of users NU.. Nevertheless, most
of time, systems are not full load but rather half load. Thus, in meantime, using
either a PIC detector or a GMMSE detector presented the same performance up to
48 users and they both outperform the SU MMSE detector. If implemented with an
adaptive algorithm, the GMMSE detector remains easier to implement than a MU
PIC detector for a lower complexity [25].
where Li depends on the mean value of the channel fading over Nc subcarriers
on which the transmitted data has been spread and zi is the symbol obtained after
despreading.
Last results include either a classical Convolutional Coding (constraint
length K = 7, polynomial generators: 133o and 177o) or a Convolutional TurboCode (constraint length K = 5, polynomial generator: 23o and 35o). The two
encoders are half rate. As mentioned in [23], decoding might be introduced into
85
86
Figure 11. BER performance with convolutional coding for a full load system,
different channels and MMSE detector.
87
Figure 12. BER performance for a half load system over a VA channel.
VI.3.3. BER performance with a Turbo Decoder versus the number of
active users
In Figure 13, performance of each detector over the BU channel is given versus
the number of users. As in Figure 9, the curves are drawn for a BER equal to 10-3
and consequently provide the necessary Eb/N0 to reach this BER.
For the BU channel, we did not carry out time interleaving. In fact, this
channel would require a 333 msec interleaver in order to fight symbol
correlation and implementing such a long interleaver is not possible for mobile
applications. Nevertheless, we can note that up to 1/3 load the GMMSE detector
allows performance close to the limit bound and is then always better than PIC
and MMSE detectors whilst the PIC detector never widely outperforms SU MMSE.
In fact, as already pointed by results in Figure 10, for a full load system, the
MMSE detector even outperforms the PIC detector. The interest of implementing a
88
VII. Conclusion
In this paper, we have presented the different Multi Carrier Spread Spectrum
systems. We have described the MC-CDMA system well adapted to the downlink
of wireless high rate system cellular networks. We have detailed the existing
Single User and Multiple User algorithms and presented a new one called
GMMSE. We have demonstrated how efficient the use of the GMMSE detector was
over BU, HT and Vehicular channels, especially when associated with a powerful
Turbo Code. Even if some system parameters may be optimized depending on
the application, the performance of MC-CDMA systems presented in this paper are
quite acceptable leading to a total spectral efficiency (including the losses due to
the insertion of pilot symbols for channel estimation) of about 0.8 bit/(sec .Hz)
for each studied channel. This spectral efficiency corresponds to 0.0125 bit/(sec .Hz)
per user in our system where 64 users can simultaneously transmit. With respect
to UMTS and IMT2000 requirements, it is possible to allocate to one user several or
all user codes such that, for a bandwidth equal to 5 MHz, a net bit rate up to 4
Mbit/sec can be assigned to a single user. A 6 Mbit/sec bit rate could be reached
with a 3/4 rate code. Thus, it is shown that MC-CDMA is a very promising multiple
access scheme especially for the downlink of future mobile radio systems.
Nevertheless, we have to keep in mind that some parameters such as channel
estimation, nonlinearity amplifiers and phase noise oscillators, as well as
synchronization devices have not yet been taken into account to complete this
analysis.
Acknowledgments
The authors, J.-F. Hlard and J.-Y. Baudais from INSA, Rennes would like to
thank FT R&D/DMR/DDH and Mitsubishi ITE who supported and contributed to this
study. Furthermore, the authors would like to express their thanks to the
anonymous reviewer for his useful suggestions.
89
REFERENCES
[1] TURIN (G.L.), Introduction to spread-spectrum antimultipath techniques and their
application to urban digital radio, Proc. of the IEEE 68, n 3, pp. 328-353,
(March 1980).
[2] PICKHOLTZ (R.L), MILSTEIN (L.B.), SCHILLING (D.L.), Spread spectrum for mobile
communications, IEEE Trans. Vehic. Tech., 40, n 2, pp. 313-322, (May 1991).
[3] LEE (W.C.Y.), Overview of cellular CDMA, IEEE Trans. Vehic. Tech., 40, n 2,
pp. 291-302, (May 1991).
[4] SHAFI (M.), Wireless communication in the twenty-first century: a perspective,
Proceedings of the IEEE. 85, n 10, pp. 1622-1638, (1997).
[5] ALARD (M.), LASSALE (R.), Principles of modulation and channel coding for digital
broadcasting for mobile receivers, EBU review, 224, pp. 3-25, (Aug. 1987).
[6] LE FLOCH (B.), HELARD (J.F.), HALBERT (R.), LEMESLE (J.M.), Modulation et codage de
canal pour la radiodiffusion sonore numrique vers les mobiles, Colloque
GRETSI'89, Juan-Les-Pins, (12-16 June 1989).
[7] ETSI ETS 300 744, Digital video broadcasting (DVB): framing structure, channel
coding and modulation for digital terrestrial television (DVB-T). (March 1997).
[8] ETSI, Project Broadband Radio Access Networks (BRAN); HIPERLAN Type 2, Technical
specification; Physical layer, (Oct. 1999).
[9] IEEE, supplement to standard for telecommunications and information exchange
between systems - LAN/MAN specific requirements - Part 11: wireless MAC and PHY
specifications: high speed physical layer in the 5 GHz band, P802.11a/D7.0,
(July 1999).
[10] CHOW (J.S.), Tu (J.C.), CIOFFI (J.M.), A discrete multitone transceiver system for
HDSL applications, IEEE JSAC, 9, pp. 895-908, (Aug. 1991).
[11] YEE (N.), LINNARTZ (J.P.), FETTWEIS (G.), Multicarrier CDMA in indoor wireless radio
networks, Proceedings of IEEE PIMRC'93, pp. 109-113, Yokohama, Japan, (1993).
[12] FAZEL (K.), PAPKE (L.), On the performance of convolutionnally-coded CDMA/OFDM
for mobile communication system, Proceedings of IEEE PIMRC'93, pp. 468-472,
Yokohama, Japan, (1993).
[13] FAZEL (K.), Performance of CCDMA/OFDM for mobile communication system,
Proceedings of IEEE ICUPC'93, pp. 975-979, Ottawa, Canada, (1993).
[14] CHOULY (A.), BRAJAL (A.), JOURDAN (S.), Orthogonal multi-carrier techniques applied
to direct sequence spread spectrum CDMA system, Proceedings of IEEE
Globecom'93, pp. 1723-1728, Houston, USA, (1993).
[15] DASILVA (V.M.), SOUSA (E.S.), Performance of orthogonal CDMA codes for quasisynchronous communication systems, Proceedings of IEEEICUPC'93,pp. 995-999,
Ottawa, Canada, (1993).
90
[16] VANDENDORPE (L.), Multitone direct sequence CDMA system in an indoor wireless
environment, Proceedings of the first IEEE symposium of communications and
vehicular technology, pp. 411-418, Delft, The Netherlands, (1993).
[17] HARA (S.), PRASAD (R.), Overview of multicarrier CDMA, IEEE Communications
Magazine, pp. 126-133, (Dec. 1997).
[18] KAISER (S), OFDM-CDMA versus CDMA: performance evaluation for fading channels,
IEEE International Conference on Communications, pp. 1722-1725 (June 1995).
[19] OH (J-Y.), LIM (M-S.), The bandwidth efficiency increasing method of multicarrier
CDMA and its performance evaluation in comparison with DS-CDMA with Rake
receivers, Proceedings of IEEE Vehicular Technology Conference, pp. 561-565,
(1999).
[20] HELARD (J.-F.), BAUDAIS (J.-Y.), CITERNE (J.), Linear MMSE detection technique for
MC-CDMA, Electronics letters, 36, n 7, pp. 665-666, (March 2000).
[21] YEE (N.), LINNARTZ (J.P.), Wiener filtering of multi-carrier CDMA in a Rayleigh fading
channel, Proceedings of IEEE PIMRc'94, 4, pp. 1344-1347, (Sept. 1994).
[22] LE GOUABLE (R.), HELARD (M.), Performance of MC-CDMA Systems in Multipath
Indoor Environments - Comparison with COFDM-TDMA System, 3G Mobile
Communication Technologies, London, (March 2000).
[23] LE GOUABLE (R.), HELARD (M.), Performance of Single and Multi-User Detection
Techniques for a MC-CDMA System over Channel Model used for HIPERLAN2,
Proceedings of IsssTA2000, New-Jersey, USA, (Sept. 2000).
[24] HELARD (J.-F), BAUDAIS (J.-Y.), CITERNE (J.), Comparaisons des performances de
diffrentes Techniques de detection Appliques un signal AMRS Porteuses
Multiples dans un Canal de Rayleigh", GRETSI'99, Vannes, France, (Sept. 1999).
[25] BAUDAIS (J.-Y.), HELARD (J.-F.), CITERNE (J.), An improved linear MMSE detection
technique for Multi-Carrier CDMA systems: comparison and combination with
interference cancellation, European Transactions on Telecommunications, Special
issue on Multi-Carrier Spread-Spectrum, 11, n 6, pp. 547-554, (Dec. 2000).
[26] MOTTIER (D.), CASTELAIN (D.), A spreading sequence allocation procedure for
MC-CDMA transmission systems, Proceedings of the VTC2000, Boston,
pp. 1270-1275, (Sept. 2000).
[27] KAISER (S), KRZYMIEN (W.A.), Performance effects of the uplink asynchronism in a
spread spectrum multi-carrier multiple access system, European Transactions on
Telecommunications, Special issue on Multi-Carrier Spread-Spectrum, 10, n 4,
(July/Aug. 1999).
[28] DASILVA (V.), SOUSA (E.S.), Multicarrier orthogonal CDMA signals for quasisynchronous communication systems, IEEE Journal on Selected Areas in
Communications, 12, pp. 842-852, (June 1994).
[29] SOUROUR (E.), NAKAGAWA (M.), Performance of orthogonal multicarrier CDMA in a
multipath fading channel, IEEE Trans. on Communications, 44, n 3, pp. 356-367,
(March 1996).
91
[30] KONDO (S.), MILSTEIN (L.B.), Performance of multicarrier DS-CDMA systems, IEEE Trans.
on Communications, 44, n 2, pp. 238-246, (Feb. 1996).
[31] PROAKIS (J.G.), Digital Communication, McGraw Hill, 1995.
[32] CORNIC (A.C.), GRANT (P.M.), THOMSON (J.S.), A Novel Interference Cancellation
Receiver for Uplink Multi-Carrier CDMA, Proceeding of ISSSTA 2000, Parsippany NJ
USA, pp. 155-158, (Sept. 2000).
[33] PATEL (P.) and HOLTZMAN (J.), Performance comparison of a DS-CDMA system using a
successive interference cancellation (Ic) scheme and a parallel IC scheme under
fading, Proceedings of IEEE Icc'94, New-Orleans, USA, (May 1994).
[34] BAUDAIS (J.-Y.), HELARD (J.-F.), Procd d'galisation dans des rcepteurs utilisant
une combinaison des techniques de modulations porteuses multiples et accs
multiple par repartition de codes. French patent n 99/11689, (Sept. 1999) and
n 99/15919, (Dec. 1999).
[35] COST 207, Digital land mobile radio communications, tech. rep., Luxembourg:
Office for Official Publications of the European communities.
[36] UMTS standard TR 101 112 V3.2.0, pp. 39-43, (April 1998).
[37] MAXEY (J.J.), ORMONDROYD (R.F.). Multi-carrier CDMA using convolutional coding
and interference cancellation over fading channels. First Workshop on MultiCarrier Spread-Spectrum, Oberpfaffenhofen, Germany, pp. 89-96, (April 1997).
[38] KAISER (S.), Multi-Carrier CDMA Mobile Radio Systems - Analysis and Optimization
of Detection, Decoding and Channel Estimation, PhD thesis, University of
Munich, Germany, (1997).
[39] BERROU (C.), GLAVIEUX (A.), THITIMAJSHIMA (P.), Near Shannon limit error-correcting
coding and decoding: turbo codes, Icc'93, Geneva, (1993).
[40] LE GOUABLE (R.), HELARD (M.), Performance of a MC-CDMA system including Turbo
Codefor wireless indoor communications. ECWT Paris, (Sept. 2000).
Chapter 4
M. Arndt
France Telecom R&D
I. Introduction
The normalization group 3GPP is preparing the standards for the third generation
of radio-mobile system, UMTS (Universal Mobile Telecommunications System).
Among them, the TDD (Time Division Duplex) mode [21] permits, for one or
several users, the simultaneous transmission, during each TDMA slot, of spread
QPSK symbols with different CDMA signatures. The uplink and downlink work at
the same carrier frequency around 2 GHz, with quadrature modulation by two
3.84 Mchip/sec-binary streams of 5 MHz-spectral width. A synchronization
channel and a midamble, transmitted at the slot rate, make the synchronization
and channel estimation easier, independently of the symbol estimation mainly
considered here. The spread spectrum systems were initially introduced in a
single-user situation in order to ensure discretion or to fight against undesired
channel effects [2]. They were calibrated, as far as possible, so that the symbol
time was superior to the temporal spread of the channel, in such a way that the
receiver was simply based on a matched filtering (MF).
Therefore, the transmission was robust but featured a very bad spectral
efficiency.
93
In contrast, the systems which are considered today should provide an increase
in the network global capacity. So, the spread messages from different users are
superimposed1, in a smaller number but as close as possible as to the spreading
factor; this results in a non-negligible part of Multiple Access Interference (MAI).
Moreover, the channel spread may go beyond the symbol time in the vehicular
environment [20], leading to substantial Inter-Symbol Interference (ISI). To
compensate interference and better detect each user's symbol, linear joint detection
with temporal depth is a receiving strategy which uses the knowledge of the codes
of other users, whereas the simple matched filter considers the MAI and IsI simply
as wideband noise. The linear joint detection is not optimal for minimizing the Bit
Error Rate (BER) but ensures a good compromise between performance and
complexity, especially in TDD mode where the codes are short (16-chip long
maximum). Nevertheless, interference cancelling, complete or partial, leads to an
amplification of thermal additive noise. This effect can be largely reduced by using
multisensor coherent processing on the mobile, which improves the waveform
orthogonality and facilitates the system inversion. We only consider non-adaptive
antenna for the base station transmitter.
The objectives of this chapter are:
(A) to interpret the CDMA system in downlink (in Section IV), usually with a
frequency approach, natural in spread spectrum systems. We will establish the
wide-band equivalent model, leading to an analogy with source separation in
antenna processing, and the symbol-band equivalent model which is identical to
the one not featuring spreading systems,
(B) to make a synthesis of the theoretical structures and performances of
"classical" linear receivers for selective channels (in Sections III, V and VI),
established from models described in (A). An exhaustive synthesis has been carried
out by the authors of [4] but from block temporal formulation using a transfer
matrix of all the symbols in the slot. The frequency approach permits here
simplification of the model and to take into account the selective channels naturally,
(C) to evaluate the average performances in Vehicular and Indoor
environment models (in Section VII) by application of the expressions of (B) in
order to measure the interest of joint detection and multi-sensor reception over
selective and non-selective channels.
94
where
- ak[.] are the QPSK symbols transmitted by the k-th source at the "symbol
times" mTs,
- gk = ck * he * h is the overall wide-band waveform of the k-th user.
It results from the convolution between the k-th spreading code, the half
Nyquist (square-root raised cosine filter "RRC" with roll-off r off ), and the
channel,
- n(t) is a baseband additive complex noise.
We consider the following context of downlink multi-user communication:
- the number of users (or active codes) K is less or equal to the spreading
factor Q;
- the K active codes (taken among a set of Q known codes) are supposed
known at the receiver, which is not necessarily the common situation
(necessity of a control channel);
- the codes of different users are made from Q binary complex entity named
"chips". The codes are linearly independent and orthogonal at zero delay
(algebraically meaning), which is expressed by:
where 5 is the Kronecker symbol whereas 5(.) is the Dirac distribution. The
impulse response of one code lasts for Ts, and is defined for the k-th user
by:
ck() =
Q-1
q=0
ck[q]
- the transmitted symbols are stationary, with zero mean and power A2; they
are uncorrelated temporally and from one user to another, and also
uncorrelated with the additive noise;
Advances in
UMTS technology
95
where L, is the number of paths, aiare the complex amplitudes and tithe
propagation delays of different paths, with zi s W, Tc,Vi{ 1, ..., Lt};
-the complex noise is an additive white Gaussian noise, with two-sided
NO for the
power spectral density 2N0 (corresponding to a two-sided psd 2
real noise on band carried at the entrance of the receiver);
- two slots of symbols are isolated by a guard time above W, * Tc ;
- we suppose perfect timing recovery (using the synchronization channel)
d -
uo
1
2 Ti,
Tcoh
. 360 around *25 during one 112 slot, in the case of a deterministic
96
On the other hand, since the mobile speed v m is always negligible compared to
the wave celerity q,,the variation of the path delay % . Tsroris negligible during
IJO
the slot (around 10-4Tc).
97
The Ts-sampled outputs {yk[m]-} are expressed from discrete time convolutions of
the transmitted symbols, disturbed by an additive noise:
where ik [n] is the sampling, at symbol time, of the continnous and deterministic
cont
H
cross-correlation function ik() = (gi * g k) () between the wide - band
waveforms of users i and k:
After matched filtering and synchronous sampling, equation (5) shows that
all the system can be modelled at discrete time Ts. The series {yik [n]}, Vn Z
providing the discrete filtering, has no dimension. Through a multi-path channel,
the waveforms are generally not orthogonal and some ISI and MAI interference
occurs with secondary terms 11 at non-zero delays and cross-channel terms 1i,
i 1 at the different delays respectively:
The necessity and the difficulty of the equalization will be entirely conditioned
by the values {yik[n]}.
Recalling that:
* In the case of a single-user system or without joint detection, only the first
branch is used, operating a coherent combination relative to the different paths,
called Rake; y1[m] may then be the decision variable.
* In the case of a multi-user system but a single-path channel, the duration
of the code limited to Ts (and respect of the Nyquist criterion at the chip time for
he) leads to no interference at non-zero delay since ik[n] = 0, Vn 0. Moreover,
when the codes are orthogonal, there is no interference at zero delay since
Yik[0] = ik: only the first branch of the MF bank is useful, reduced to the receiving
filter RRC followed by correlation with the desired code.
98
Having described the structure of the detector, we will now derive and
interpret equivalent models of the CDMA system before setting the performances
and expressions of the multi-user equalizer.
hwit
99
where by convention:
-fw is used for any frequency while / is a frequency of the symbol-band,
included in [1\2Ts,+1\2Ts].So, for all fw there is a single frequency / and
a single relative integer q such that fw = f + q\Ts,
- the continuous Fourier transform of a function h in continuous time or
discrete time (then obtained via Z transform) is noted h(.).
The transmission bandwidth, defined by the support gk(fw) is limited to
the band of the half Nyquist filter, that is Qr.1\Tswhere Qr is the integer part of
(1 + roff ). Q, expressing the filter bandwidth excess. With the UMTS parameters
Q = 16 and roff = 0.22, Qr is odd and equal to 19.
The Qr symbol bands (of width1\Ts)which comprise the wide-band are then
centred around the frequencies fw = Qr/2\Ts,..., 0, ..., +Qr/2\Tswith the definition of
Qr/2 such that Qr = 2Qr/2 + 1.
Note: Taking into account the RRC attenuation, the portions not included in
this Qr bands are ignored. If one does not want to lose any information, we might
have taken Qr + 1 = 20 instead of Qr and choose for f a range equal to [0, + ].
Due to the characteristic of "sampling" (cf (10)) of ak(t) at the symbol rate
(or the cyclostationarity of the digital communication signals), k(fw) is periodic
with a period of 1\Ts:
stream akrect(t) (shaped into a rectangular pulse of length Ts) by the periodic
version of the code ckper(t) (with period Ts). The expression of the periodic code
in Fourier series gives then the multi-carrier modulation scheme directly. This
other equivalent representation of the spreading system, i.e. the "multiplicative
representation", is closer to the design, and probably more traditional [3].
However, it is more difficult to manipulate theoretically than the representation
with linear filters, generally adopted.
IV.2. Interpretation at reception
Assuming a perfect knowledge of the channel and timing synchronization, the
outputs of the bank of MF-samplers can be synthesized from a linear
transformation of the transmitted symbols, without mentioning the wide-band
received signal. The temporal expression, at discrete time Ts, of the "symbolband model" has been established in (5).
The transfer yik(f) between the symbols of the source "i" and the output of
the branch "k" is expressed in frequency-domain by Fourier transform of
equation (6) as the aliasing of the wide-band cross-spectrum y ik c nt (f w ) in the
symbol-band:
with:
Respect of the first Nyquist criterion, relative to the chip time Tc for he * he H,
leads then to:
For the desired user, equation (19) simply expresses the first Nyquist
criterion relative to the symbol time Ts for g1 * g1 H with an ideal channel. When
the channel is not single-path, the aliasing equation (16) explains the robustness
of the spread-spectrum to selective channels: the aT/RGC seen by the desired user
symbols is finally a channel with width obtained from the coherent
75
superposition of Qr channels with equal width, averaging the selectivity imposed
to the different bands.
Figure 3 shows the T/RGC and aT/RGC on branch one for the desired user and
for the interferer "2", with a channel taken from the Vehicular B model,
The power of IsI is derived from the quadratic frequency area between the
desired aT/RGC and an ideal flat (white) one: it results thus from the
non-whiteness of the desired aT/RGC. The power of MAI is derived from the sum
of the quadratic frequency area of interfering aT/RGC relative to the branch "1".
The noise power is derived from the Noise Equivalent Bandwidth: Y11[10] . 1\Ts of
the branch "1", i.e. from the frequency area in the symbol band of the desired
aT/RGC.
where
The beamforming favours one spatial direction for extracting the source "1"
from the coherent recombining of the outputs of the Qr sensors. In the same way
in CDMA, the matched filter favours one "frequency signature" by coherently (in
branch "1" for the desired user) re-combining the components of the Qr bands
or sub-carriers, which appears in (4) and can be formulated as:
where
where
Two examples of the matrix | (f) | are given for K = 8 users (with linear
scales): in Figure 6, the temporal spreading of the channel (one realization of the
Pedestrian B channel according to the ETSI model) is around 1 75.
In Figure 7, the temporal spreading is around 5 Ts (same Vehicular B channel
as used in Figure 3). In the two examples, the interference is around the same
order but we can verify that the coherence bandwidth in Vehicular B is about 5
times smaller.
Advances in
where 1'
2
-
UMTS
technology 109
In the absence of ISI and MAI, the matched filter obtains the best
2E
to (>)
as well as the ideal elementary Binary Error Probability.
NO
SINR
equal
(MSE>& E{ 1 d,,,]
a,[,] 1 2>
The integration term in this expression is the power spectral density of the
error (psd), comprised of a ISI term, a MAI term, and a noise term.
From matrix formulation:
- the vector t?, (f) = [e^,,cf), ..., gIKcf)] is created from the transfer
functions ofthe equalizing filters on the K branches,
-
the global transfer function for the symbols of the user k is: giolkcf)= e^,-
<ne E 0 - k
- the global transfer function t1(f) for the noise (with periodicity due to
the sampling) is such that:
Notes:
* Interference power plus noise can be obtained simply from the MSE by
subtracting the quadratic error term at zero delay: (IN) E q = (MSE) E q
A2 \ l-glo 1 1 [ 0 ] \2.
* The global linear detector 11 for the detection of the symbols of user "1",
including the MF-sampler bank and the equalizing filter bank, has a transfer
function l1(fw) defined for the wide-band. It achieves actually, for one frequency
f of the symbol-band, a linear combination of the Qr sub-bands. Its expression
under vector form of size Qr (grouping l 1 ( f - Qn/2\Ts).... l1(f + Qn/2\Ts), is:
V.2.1. ZF equalizer
The optimal ZF equalizer minimizes the MSE, under the constraint of cancelling
the ISI and MAI. With such a criterion, in situations without noise, the estimated
symbols must be equal to the transmitted symbols for the desired user. At the
same time, among all the wide-band global filters l1 which cancel the
interference, this criterion warrants the one with a minimal norm and leads to the
structure in Figure 2 (or equation (27)). The cancelling of the interference from
the symbol-band model requires then that glolk(f) = k1, which is equivalent to
glo1k[n] = k1n in discrete time representation.
The equalizer vector is then obtained by inverting the spectral matrix of
aliased cross-channels:
The zero delay coefficient of the equalizer in branch "1", e11zf[0], is a posi
real number, derived from the diagonals of the Hermitian matrix. It plays a
fundamental role since it describes entirely the performance of the system. More
precisely, it expresses a degradation factor for the SINR but also for the Error
Probability (Pe or BER)2 since after perfect theoretical equalization, we obtain
exactly the desired symbols, only disturbed by a gaussian noise:
- 1
/ example in Pedestrian B.
situation, the partial inversion of the MMSE limits the amplification of the noise to
the frequencies where the local signal to noise ratio is bad. Moreover, since the
power of the noise is not zero, the inverse of '=(f) always exists an ill conditioned
situation is impossible, contrary to the ZF case.
The global gain (at zero delay) of all the chain Tx/Rx for user "1 ",glo11me[0]
=
a noise-free situation). The MMSE receiver has the same behavior as the ZF
Eb
receiver in situations of strongEb/N0and is reduced to the receiving head, i.e. to
the MF, inthe extreme opposite situation. The zero delay coefficient of the
equalizer is here normalized by the global gain to form the degradation factor
of the SINR ( 1), which is expressed through:
In the presence of selective aT/RGC in the symbol band, the equalizing filters
may use practically a finite temporal depth of P coefficients (with causal and
anti-causal parts). We then have added the frequency dimension /, which is
equivalent to independently treating several non-selective channels, with narrow
bands compared to (second order stationarity).
This approach naturally (but approximately) provides the coefficients of the
equalizer by using the Inverse Discrete Fourier Transform, and constitutes an
alternative solution to temporal methods [12, 4], with little complexity.
To obtain a temporal depth of P coefficients for the equalizer, we just have
to sample the symbol band with a step f =1/PTsto form the P matrix=(p f)
and compute their inverse. The K non-causal filters {e l k [ p ] } used to detect the
symbols of user "1" are obtained by a DFT - 1 of the K spectral patterns of the
line "1" of the inverse matrix. The filter {e 1 1 [ p ] } born from the diagonal is the
most energetic and corresponds to the main branch. In a multi-code case, i.e.
with several desired codes, we can take advantage of the hermitian symmetry of
the taps: eik[n} = e*ki[_n], Vi, k = 1...K, Vn ZZ.
Notes:
* The impulse responses of the equalizer bank at symbol time after the MF
bank are theoretically infinite. So, the choice of P should ensure a negligible
temporal aliasing for the practical filters. Depending on to the weak values of the
secondary far paths in the ETSI models described in Section VII, a typical
temporal depth of 1 to 4 coefficients in indoor A and of 8 to 16 coefficients in
Vehicular B is generally sufficient.
* Instead of preserving approximately the theoretical imposed structure
described in Figure 2 (with a finite number of taps for the practical equalizer), it
is also possible to compact the linear detector l1 in a single discrete transverse
fractional filter with a cadence of Tc/2 at input and Ts at output.
This code "1" is then reduced to the complex version of the scrambling code.
VII. 1.2. Propagation channels
Models described by ETSI [20] in "Indoor A" (IA) and "Vehicular B" (VB)
environments are used. The channel impulse responses are constituted by 6 paths
with relative delays (to the first paths) and average power, summarized in the
following table:
IA (nsec)
50
110
170
290
310
(dB)
-3
-10
-18
-26
-32
VB (nsec) 0
0.3
8.9
12.9
17.1
20.0
-13
-10
-25
-16
(dB)
-2.5
Advances in
UMTS technology
117
The temporal spreading is very weak in IA, slightly superior to one chip time,
Tc; on the contrary, the temporal spread is important in VB, around 5 23, which
will permit equalizer performances in two very different situations. The
propagation delays are quasi-fixed (variation of % 3%). The complex
amplitudes a: of the paths are random, distributed as with a Rayleigh law for the
modulus and uniform law between 0 and 2n for the phases. We suppose the
complex amplitudes uncorrelated between two paths (index i) and from one
sensor to another (index l). This last hypothesis, more often admitted [lo], is
certainly a bit optimistic for the inter-sensor spacing looking at the mobile; we
will come back to this point in the performance analysis.
(SINR)
1
(3)
- where PI
No
$1
represents the degradation on the SINR caused by interference for the Matched
Filter, Zero-Forcing and MMSE receiver. For the Zero-Forcing equalizer, PI also
represents the degradation of the BER.
We recall that these performance formulae suppose perfect knowledge of the
channel and perfect timing synchronization, they give then upper bounds which
will inevitably be degraded in real situations. Moreover, they are only valid for
a deterministic and time invariant impulse response, which does not correspond
to IA and VB models since they are random, mainly from the complex amplitudes
a:. Rather than calculating the performances for one specific observed sample
of the path amplitudes (supposed then deterministic, equal for example to their
quadratic average values), we are going to give average performances for the
whole possible occurence of impulse response. This classical process [17, 101
may be used to represent a practical situation with slowly time variant channel:
one occurence of af is associated with the channel impulse response at a given
instant. The channel variation is slow enough to be considered fixed during the
occurence and perfectly estimated. So, the interference term PI, as with the
energetic gain of the global waveform Y,,,~, become random variables. We are
now interested in the expected values of the SINR inverse (named INSR) and of
the Error Probability (for a),
computed from a very large sampling number:
where
0,
yll[ol =
Yll[OI
Eb
&,
13,
No
- L.
The slope of
the BER curve (with classical Logscales) is imposed by the diversity order, equal
to the number of sensors in this basic example.
With any random channel, we may still define an equivalent diversity order
Neq relative to a Rayleigh fading, as proposed in [9], from the variation
coefficient cv of the power received:
We then dispose of two indexes to measure the depth of the fading: cv and f
With respect to a Rayleigh fading with N independent diversity branches (with
same power), it can be shown that:
Neq
cv (dB)
f(dB)
-1.5
-2.3
-3.5
-4.5
00
1.76
0.96
0.57
+ 00
+ 00
matched filter. For the MF, the (INSR) curves reach a floor, for strongEb/N0and for
the large values of K, because of the interference. The improvement due to the
equalization is emphasized with 2 or 3 sensors, which brings it almost back to
ideal performances.
Concerning the BER in ZF, there is an important loss, around 15 dB at
Pe = 10-3 when the saturation of the number of users (K = 16) is approached.
The multisensor reception gives a new degree of freedom and reduces the
degradation to just 1 dB with 3 sensors.
VIII. Conclusion
This study has permitted, by taking a frequency interpretations approach, to
recall non-exhaustively the linear detection structures in the downlink of a TDCDMA system. We have formulated the expressions and performances of this
detector from the representation in the symbol-band, based on the spectral
matrix of the aliased cross-channels. Finally, we have applied the performances
expressions to the VB and IA environment models. These performances constitute
upper bounds for real situations because of the inaccuracy in channel estimation
and in synchronization operation.
From the VB model, frequency selective, we have first illustrated the benefit
of the joint detection as compared with the simple desired waveform matched
filter, especially when the code margin is little. We have also underlined the clear
performance improvement for a coherent reception with 2 or 3 sensors. In this
selective channel, the multi-sensor reception permits the decrease of interference
and to facilitation of the inversion performed by the equalizer.
From the IA model, weakly selective but also very poor in temporal diversity
(non-resolved paths), we notice a spectacular improvement of performances due
to the diversity gain of the multi-sensor, reducing the range of amplitude fading.
So, multi-sensor reception on the mobile is always recommended to fight
efficiently the two different types of distortion introduced by the channel:
frequency selectivity and fading process.
REFERENCES
[1] VAN ETTEN (W.), Maximum likelihood receiver for multiple channel transmission
systems. IEEE Transactions on Communications, pp. 276-283, (Feb. 1976).
[2] TURIN (G.L.), Introduction to spread-spectrum antimultipath techniques and their
application to urban digital radio, Proceedings of the IEEE, vol. 68, n 3,
pp. 328-353, (March 1980).
[3] PICKHOLTZ (R.L.), SCHILLING (D.L), MILSTEIN (L.B.), Theory of spread-spectrum
communications - A tutorial. IEEE Transactions on Communications, vol.com-30,
n 5, pp. 855-884, (May 1982).
[4] JUNG (P.), BLANZ (J.), Joint detection with coherent receiver antenna diversity in
CDMA mobile radio systems. IEEE Transactions on Vehicular Technology, vol. 44,
n 1, pp. 76-88, (Feb. 1995).
[5] MADHOW (U.), HONIG (M.L), MMSE interference suppression for direct sequence
spread-spectrum CDMA. IEEE Transactions on Communications, vol. 42, n 12,
pp. 3178-3188, (Dec. 1994).
[6] WANG (X.), POOR (H.V.), Blind multiuser detection: a subspace approach.
IEEE Transactions on Information Theory, vol. 44, n 2, pp. 677-690, (March 1998).
[7] BALABAN (P.), SALZ (J.), Optimum diversity combining and equalization in digital
data transmission with applications to cellular mobile radio-Part I: Theorical
considerations. IEEE Transactions on Communications, vol. 40, n 5, pp. 885-894,
(May 1992).
[8] BOURENNANE (S.), FAURE (B.), L.ACOUME (J.L.), Traitement d'antenne pour des sources
bande large. Annales des Tlcommunications, vol. 45, n 7-8, pp. 384-392,
(July 1990).
[9] DIOURIS (J.F.), ZEIDLER (J.), BULJORE (S.), Space-path diversity in CDMA using a compact
array. Annals of Telecommunications, vol. 53, n 11-12, pp. 425-434, (1998).
[10] BULJORE (S.), DIOURIS (J.F.), SAILLARD (J.), tude thorique d'un galiseur multicapteur associ au critre EQM pour le canal radiomobile (modle GSM),
GRETSI, Juan-les-Pins, pp. 557-560, (Sept. 1995).
[11] VILA (P.), PIPON (F.), PIREZ (D.), FETY (L), Filtrage adapt multidimensioonel pour
l'galisation d'un canal slectif en frequence et brouill. GRETSI, Juan-les-Pins,
pp. 553-556, (Sept. 1995).
[12] PICEONNAT (Y.), Alternative solutions for joint detection in TD/CDMA multiple access
scheme for UMTS. SPAWC Anapolis, pp. 329-332, (May 1999).
[13] Ros (L), JOURDAIN (G.), ARNDT (M.), Multi-sensor reception for TDD-UMTS mobile
terminal in indoor and pedestrian environments. 11th IEEE International
Symposium on PIMRC, London, pp. 1221-1225, (Sept. 2000).
[14] Ros (L.), Reception multi-capteur pour un terminal mobile TDD-UMTS. Report us
n8!2000, (May 2000).
Advances in
UMTS technology
125
PRESS,
New-York, (1974).
UMTS 30.03,
TR101-112
(March 2000).
Chapter 5
Smart-antenna space-time
UMTS uplink processing for
system capacity enhancement
T. Neubauer
Technical University, Vienna, Austria
E. Bonek
Technical University and FTW, Vienna, Austria
I. Introduction
The European third generation mobile communication system is known as the
Universal Mobile Telecommunications System (UMTS). It will enable tomorrow's
wireless Information Society, to deliver high-value broadband information,
commercial and entertainment services to mobile users. UMTS will play a key
role in creating the future mass market for high-quality multi-media
communications that will approach 2 billion users worldwide by the year 2010
[1].
The main advantage of third generation mobile systems will be mixed
services with variable data rates. The customer profile of a 3G system will
consist of a large number of low-data-rate and voice users with the concurrent
service to a high-data-rate user population.
The frequency division duplex (FDD) mode of UMTS is based on a Wideband
Code Division Multiple Access (W-CDMA) technique. CDMA systems are
interference limited in nature and suffer severely from the near-far effects
because all users operate in the same frequency band simultaneously. This is
essentially more critical in the uplink of UMTS where all users are potential
interferers. The problems due to the near-far effect are reduced in the downlink
because of the better orthogonality of the individual signals [2].
Since higher bit-rate services experience lower processing gains, they have
to be sent with higher power levels in order to fulfil the quality requirements at
the receiver. This causes greater interference at the base station (BS) antenna and
an intensification of the near-far problem. Hence, the capacity will be reduced
due to the overall interference increase.
It is well accepted that the received signal at the BS originates from scatterers
within a disk of radius r centred R metres from the BS around the UE [24].
Measurements in various urban environments have shown that the spatial
behaviour of the incoming signals at the BS can be modelled by a main direction
of arrival (DOA) in the azimuth and a corresponding angular spread [5].
The path loss is modelled based on the proposed models for UMTS system
simulations presented in [2] for pre-defined environments. Path loss correlation
between two up-link connections from a single UE to different BSS is expressed
by a correlation coefficient of 0.5 [10]. The log-normal large-scale fading is
2
modeled with a mean of In = 0 and a standard deviation of ln = 6 dB.
II. 1.2. Spatio-temporal channel model
Within the European research initiative COST 259 [4] several channel models
have been developed. They are aimed at UMTS and HIPERLAN, with particular
emphasis on adaptive antennas and directional channels. They have been
introduced in the third generation standardization process by 3GPP [3], and they
are used here in the following way.3
The path loss again is modelled based on the proposed models for UMTS
system simulations presented in [2] for pre-defined environments. Path loss
correlation and the log-normal large-scale fading are modelled in the same way
as in the flat fading case.
In order to test advanced antenna array systems, the main focus of the
channel model has to be on the azimuthal and temporal properties. The statistics
of the delay and azimuth of the impinging waves as well as their expected power
are especially of great interest. According to [4] and [5] we define a Laplacian
shape for the Azimuth Power Spectrum (APS) seen at the BS antenna. For the
Delay Power Spectrum (DPS) we assume the widely used one-sided exponential
decaying function [5]. Both can be written as
where , , t and denote the azimuth towards the UE seen at the BS, the azimuth
spread, the delay and the delay spread, respectively. For typical urban
environments the azimuth power delay spectrum can be written as
2. Enlarging the standard deviation to ln = +10 dB has negligible impact on the results.
3. Within 3GPP [3] only thee temporal domain is considered till now.
In our model we assume a tapped delay line model for a typical urban
scenario. We consider a single scattering cluster and L impinging multipathcomponents. The delay spread is = 0.5 sec and the start of the cluster is at
0 = 0 sec [3]. For the azimuth domain we assume a spread of a, = 8 around
0, where 0 corresponds to the azimuth toward the UE. In Figure 2 both APS a
DPS are shown schematically and for the parameters above.
57
Inter-BS distance
1000m
1n = 0,1n = 6dB
0.5
Max. UE TX power
21 dBm
80 dB
= 0, p= 0.5dB
1 and 4
d = /2
10
= 0.5sec
= 8
where V Pl,k,, and 9l,k, denote the amplitude and the phase of the l-th received
signal path at a single antenna element impinging with delay l,k, and azimuth
l,k,.(,) is the Dirac delta function. The array response vector for a single
UE k at BS , for the delay is thus given as
where cl,k,, is the array steering vector of a wave impinging from an azimuth
direction l,k, at an M-element uniform linear array with an inter-element
spacing d.
In the FDD mode of UMTS, the time a wavefront takes to pass through the array
is much smaller than the chip interval Tc. Therefore, the narrowband assumption
for antenna arrays is valid. This makes it possible to model the time delays as
phase shifts. The array steering vector can thus be written as
In the same way we define the spatial interference and noise covariance
where 2N and I denote the noise variance and M x M the identity matrix,
respectively.
In the case of the flat fading channel model, we include an angular spread.
Borrowing the technique from [25], we derive the spatial covariance matrix of
the k-th user at a certain instant in time as
with
Without making any assumptions about the spatial covariance matrix behaviour,
the solution for this problem is to take the generalized eigenvector
corresponding to the largest eigenvalue [R(s)k, Q(s)k,] [15]. This well-known
beamforming concept corresponds to maximizing the signal-to-noise-plusinterference ratio SNIR in the uplink [14] and the estimated ratio of signal to
interference power for the k-th UE. From
we obtain the optimum spatial weights for the k-th UE in cell as a function of
the delay . Since R contains all signal information, the received signal power
after spatial processing for the n-th delay tap is given by
where hn(T) and ()* denote the temporal channel impulse response and complex
conjugate, respectively.
IV. Simulations
For investigations of system behaviour, we now have to consider a number of
cells. As already mentioned, we consider 19 3-sectored sites, leading to a total
of 57 cells. In doing so, we correctly estimate the inter-cell interference at the
inner site and thus we are able to come up with reliable conclusions at the end
[9].
Taking the whole network into account, we calculate the interference seen at
the BS antenna for each user. Since we now have a total number of K users in the
entire network, we easily find from (8)
4. Two adjacent taps are usually not completely uncorrelated due to filtering and oversampling.
Analysis in [16, chapter 29.3 "The selective rake receiver"] has shown that the correlation is in the
range of 0.02 for raised cosine pulse shapes. Thus, we assume two adjacent taps to be uncorrelated.
the spatial covariance matrix including the total network interference for the
k-th user belonging to cell . Following the description above and repeating the
procedure for = 1, ..., Z (Z = 57) cells, we can finally calculate the quality
requirement at the BS antenna connector for each user in the network.
Introducing the processing gain GP as a function of the service of the k-th
user
with the modulation bandwidth W and the user data rate Rk, , we obtain the ratio
of combined received energy per information bit to the effective noise power
spectral density, Neff, at a single point in time.
RSCP k,
and
ISCP k ,
Interference Signal Code Power (including noise) of the k-th user in cell after
beam forming and rake combining.
For the computer simulation of the above model, we use a stationary MonteCarlo approach [9, 22, 23]. The schematic flowchart of the system simulator is
shown in Figure 7.
The main simulation parameters are given in Table I.
Figure 7.
Simplified flowchart of the
system simulator.
Figure 8 shows the relationship between uplink noise rise seen at the BS antenna
and the system load in %. The term is defined as
where SIRloaded and SIR empty are the signal-to-interference-ratio at the BS antenna
in a loaded cell (or system) and in an empty cell (or system), respectively [22].
V.1.2. Space-only results
We consider the flat fading propagation channel as described in Section 2. The
individual multi-path components receive at = 0 at the BS antenna.
We now make a comparative performance evaluation of simplified multiservice scenarios of the UMTS FDD uplink. We consider two systems, one equipped
with single-element antennas and the other with four-element antenna arrays at
all base stations within the network. Further we assume a service mix of speech
users with a net data rate of 12.2 kbit/sec and 64 kbit/sec data users. As
schematically explained in Figure 1, we expect that the gain of smart-antennas
over the single-element antenna is higher, if the percentage of high-data-rate
users in the network is high. Hence, we compare the noise rise of both systems
for various service mixes, starting with speech users only.
We can observe from Figure 9 that the shape of the noise rise in case of
speech users only is a smooth one for both systems. Taking a closer look, e.g.,
at the noise rise at the 50 users/site-level, we find a value of 5.8 dB for the nonadaptive-antenna system, while the noise rise for the smart-antenna network is
only about 2.8 dB. This 3 dB gain in the noise rise is in the range of what was
expected for CDMA systems with speech users only [18, 19].
The noise rise is increasing when we add more users to the single-antennaelement system. In contrast, the noise in the smart-antenna network rises only
slightly as a function of both the number of speech users and the increased
number of 64 kbit/sec-data users.
CDMA systems are usually planned to operate at a system load of = 50%
(as defined in Equation 19, [22]). For this operation point, Figure 10 defines the
plane of possible business cases, for both systems. A business case is the number
of users for the different services in the system area. The dashed lines (with the
marker for the single antenna case, without marker for the smart-antenna case)
indicate the respective capacity maximum, i.e. for 50% of system loading. To the
left of these lines, we can look for viable business cases. We can obtain an
ensemble of possible user ratios in our system. One such user ratio (a "business
case") with 54% data users will serve a total of 15 users (8 64 kbit/sec data, and
7 12.2 kbit/sec speech) in the conventional system, and about 48 users (26-64
kbit/sec data and 22 12.2 kbit/sec speech) by applying smart-antennas to the Bss.
The actual capacity gain, of course, depends on individual realization of the
network.
This directly leads to the question of the effective throughput as a function of
the system load and the percentage of high data rate users in the system.
Figure 9. Noise rise
as a function of both
the service mix of
speech (12.2 kbit/sec)
and 64 kbit/sec data
users and the number
of users per site.
Two curves are
shown. One for the
noise rise for the
system using
4-element antenna
arrays, and a second
one for the single
element antenna.
The latter completely
overlays the first one.
Figure 10.
Boundaries for
a single-element
antenna system
(lower left area) and
for a smart-antenna
system (upper left area)
as a function of the
system load.
The boundary is shown
for a system load of
= 50%. Arrows
indicate a specific
business case
(54% data users).
As indicated in Figure 11, the maximum effective throughput for individual
business case realizations depends on the system load as well as on the service
mix. For smart-antenna systems, the maximum throughput is steadily increasing
with the number of high-data-rate users, but that is not the case for conventional
antenna systems. This fact will be a main challenge for future UMTS operators
and network planners. They have to optimize the network depending on the
requirements of an increasing number of future data users, as well as on the
system performance.
The plateaus, the kinks and bumps, are a consequence of our taking the 95%
percentile value of CDFS.
Figure 12. Noise rise vs. the number of users per cell for aflat fading and
the proposed spatio-temporal channel model.
Figure 13. Noise rise in the central cells. The BS are equipped with
1,4 or 6 antenna elements and the temporal rake receiver combines 2 or 4 taps.
Future UMTS base stations will provide at least four rake fingers per receiver
branch. Thus, the flat fading assumption turns out to be a lower bound for the
capacity estimation for both conventional and smart-antenna systems,
respectively.
VI. Conclusions
The main focus of this study is the benefit of smart-antennas in the uplink of a
multi-service UMTS FDD system. We demonstrated the importance of arealistic
channel model accounting for spatio-temporal characteristics of mobile radio
channels for a fair assessment of smart-antenna benefits.
We applied the capability of antenna array processing to our stationary
Monte-Carlo UMTS system simulation approach. Based on this we investigated
the mixed service capacity in the uplink. We compared and evaluated the
performance of space-only and space-time processing under both flat fading and
spatio-temporal channel conditions. We investigated systems with singleelement antennas, four- and six-element antenna arrays at all base stations within
the network. Also we looked at the effect of a limited number of rake fingers in
an urban environment. Building on the results in Section 5 we conclude:
There can be a substantial increase in system capacity by incorporating
antenna arrays at the BS.
The higher the number of high data rate users is, the more will smartantennas outperform conventional antenna systems.
Acknowledgement
The authors would like to thank Klaus Hugl, Werner Weichselberger and Thomas
Baumgartner for fruitful discussions. They also acknowledge the useful editorial
comments by Martin Tltsch and the reviewers of this paper.
REFERENCES
[1] UMTS-Forum, "The Future Mobile Market", Report n 8, www.umts-forum.org,
(March 1999).
[2] 3gpp, "RF System Scenarios -TS 25.942", http://www.3gpp.org, (April 2000).
[3] 3GPP, "Deployment aspects - TS 25.943", http://www.3gpp.org, (April 2000).
[4] CORREIA (L.H.), ed., Wireless Flexible Personalized Communications - COST 259:
European Co-operation in Mobile Radio Research. J. Wiley & Sons, (March 2001).
[5] PEDERSEN (K.I.), MOGENSEN (P.E.), FLEURY (B.H.)," A stochastic model of the temporal
and azimuthal dispersion seen at the base station in outdoor propagation
environments", IEEE Transactions on Vehicular Technology, 49, n 2, pp. 437-447,
(March 2000).
[6] BELLO (P.A.), "Characterization of Randomly Time-Variant Linear Channels",
IEEE Trans. on Communication Systems, CS-11, pp. 360-393, (Dec. 1963).
[7] DELL'ANNA (M.), AGHVAMI (A.H.), "Performance of optimum and suboptimum
combining at the antenna array of a W-CDMA system", IEEE Journal on Selected
Areas in Communications, 17 n 12, pp. 2123 -2137, (Dec. 1999).
[8] NAGUIB (A.F.), PAULRAJ (A.), "Performance of wireless CDMA with M-ary orthogonal
modulation and cell site antenna arrays", IEEE Journal on Selected Areas in
Communications, 14, n 9, pp. 1770 -1783, (Dec. 1996).
[9] NEUBAUER (T.), BAUMGARTNER (T.), BONEK (E.), "Necessary and Sufficient Network Size
for Pole Capacity Estimation in UMTS FDD", ECWT 2000, Paris, France, pp. 131-134,
(Oct. 2000).
[10] VITERBI (A.J.), VITERBI (A.M.) "Other-Cell Interference in Cellular Power-Controlled
CDMA", IEEE Transactions on Communications, 42, n 2/3/4, pp. 1501-1504,
(Feb.
1994).
[11] KOPSA (K.), WEINMANN (R.), BRAUN (V.), TANGEMANN (M.), "Performance of Multiple
Antenna Combining Algorithms in the Uplink of UMTS/FDD", Globecom 2000-11-27.
[12] A. PAULRAJ (A.), PAPADIAS (C.B.), "Space-time processing for wireless
communications", IEEE Signal Processing Mag., 14, pp. 49-83, (Nov. 1997).
[13] THOMSON (J.), GRANT (P.), MULGREW (B.), "Smart antenna arrays for CDMA systems",
IEEE Personal Commun. Mag., pp. 16-25, (Oct. 1996).
[14] WINTERS (J.H.), "Optimum Combining in Digital Mobile Radio with Co-channel
Interference", IEEE Journal on Selected Areas in Comm., SAC-2, n 4,
pp. 528-539, (1984).
[15] GERLACH (D.), PAULRAJ (A.), "Adaptive transmitting antenna methods for
multipath environments ", IEEE Global Telecommunications Conference,
GLOBECOM '94, 1, pp. 425-429, (1994).
[16] MOLISCH (A.F.), "Wideband Wireless Digital Communications", Prentice-Hall,
(2000).
[17] KUCHAR (A.), TAFERNER (M.), TANGEMANN (M.), HOEK (C.) "Field trial with a
GSM/DCS1800 smart antenna base station" Vehicular Technology Conference,
Amsterdam, Holland, 1, pp. 42-46, (1999).
[18] LIBERTI (J.C.), RAPPAPORT (T.S.), "Analytical Results for Capacity Improvements in
CDMA", IEEE Trans. on VT, 43, n 3, pp. 680-690, (Aug. 1994).
[19] NAGUIB (A.F.), PAULRAJ (A.), KAILATH (T.), "Capacity Improvement with
Base-Station Antenna Arrays in Cellular CDMA", IEEE Trans. on VT, 43,
pp. 691-698, (Aug. 1994).
[20] HAGERMAN (B.), MAZUR (S.), "Adaptive Antennas in is-136 Systems", IEEE 48th
Vehicular Technology Conference 1998, Ottawa, Canada, pp. 2282-2286,
(May 1998).
[21] BORMANN (R.), FRERICH (M.), AHRENS (F.), HENB (T.), DAM (H.), BERG (M.), ANDERSSON (S.),
ALMQVIST (D.), "Functional Tests of adaptive antenna base stations for GSM",
3rd EPMCC 1999, Paris, France, pp. 323-327, (March 1999).
[22] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), JSBERG (M.), "Static simulator for
studying wcdma radio network planning issues", IEEE 49th Vehicular Technology
Conference 1999, VTC' 99 spring, Houston, USA, (May 1999).
[23] DEHGHAN (S.), LISTER (D.), OWEN (R.), JONES (P.), "W-CDMA capacity and planning
issues", IEE Electronics & Communication Engineering Journal, pp. 101-118,
(June 2000).
[24] FUHL (J.), MOLISCH (A.F.), BONEK (E.), "Unified channel model for mobile radio
systems with smart antennas", IEE Proceedings - Radar, Sonar and Navigation,
145, n 1, pp. 32- 41, (Feb. 1998).
[25] RIBA (J.), GOLDBERG (J.), VAZQUEZ (G.), "Robust Beamforming for Interference
Rejection in Mobile Communications", IEEE Trans. on Signal Processing, 45, n 1,
pp. 271-275, (Jan. 1997).
[26] LEE (J.S.), MILLER (L.E.), "CDMA Systems Engineering Handbook", Artech House Inc.,
(1998).
[27] BRUNNER (C), HAARDT (M.), NOSSEK (J.A.), "On space-time RAKE receiver structures
for WCDMA", Conference Record of the Thirty-Third Asilomar Conference on
Signals, Systems, and Computers, 2, pp. 1546-1551, (1999).
[28] KHALAJ (B.H.), PAULRAJ (A.), KAILATH (T.), "2D RAKE receivers for CDMA cellular
systems", IEEE Global Telecommunications Conference, GLOBECOM '94, 1,
pp. 400-404, (1994).
[29] WIN (M.Z.), CHRISIKOS (G.), MOLISCH (A.F.), SOLLENBERGER (N.R.), "Selective Rake Diversity
in Multipath Fading with Arbitrary Power Delay Profile", IEEE Global
Telecommunications Conference, GLOBECOM 2000, San Francisco, (27 Nov.-1 Dec.
2000).
[30] PROAKIS (J.G.), "Digital Communications", McGraw Hill Book Comp. Inc.,
3rd edition. New York, (1995).
[31] PEDERSEN (K.I.), "Antenna Arrays in Mobile Communications", Ph.D. Thesis,
Center for PersonKommunikation, Aalborg University, Denmark, (Jan. 2000).
[32] LANCASTER (P.), TISMENETSKY (M.), "The Theory of Matrices", Second Edition,
Academic Press, (1985).
[33] CUPO (R.L.), GOLDEN (G.D.), MARTIN (C.C.), SHERMAN (K.L.), SOLLENBERGER (N.R.),
WINTERS (J.H.), WOLNIANSKY (P.W.), "A four-element adaptive antenna array for
is-136 PCS base stations" IEEE 47th Vehicular Technology Conference, 3, (1997).
[34] ADACHI (F.), SAWAHASHI (M.), SUDA (H.), "W-CDMA: Performance evaluation and
future enhancement", 3rd European Personal Mobile Communications
Conference, EPMCC, Paris, 9-11, pp. 429-434, (March 1999).
[35] 3Gpp, "Vocabulary for 3gpp Specifications - TS 21.905", http://www.3gpp.org,
(April 2000).
[36] NEUBAUER (T.), JGER (H.), FUHL (J.), BONEK (E.), "Measurement of the Background
Noise Floor in the UMTS FDD Uplink Band", 4th European Personal Mobile
Communications Conference, EPMCC, Vienna, (Feb. 2001).
Chapter 6
I. Introduction
As the launch of third generation technology approaches, operators are forming
strategies for the deployment of their networks. These strategies must be
supported by realistic business plans both in terms of future service demand
estimates and the requirement for investment in network infrastructure.
Evaluating the requirement for network infrastructure can be achieved using
system dimensioning tools capable of assessing both the radio access and the
core network components. Having found an attractive business opportunity,
system deployment must be preceded by careful network planning. The network
planning tool must be capable of accurately modelling the system behaviour
when loaded with the expected traffic profile. The third generation cellular
systems will offer services well beyond the capabilities of today's networks. The
traffic profile, as well as the radio access technology itself, form the two most
significant challenges when dimensioning and planning a W-CDMA based third
generation system. The traffic profile describes the mixture of services being
used by the population of users. There are also specific system functionalities
which must be modelled including fast power control and soft handover. In order
to accurately predict the radio coverage the system features associated with
W-CDMA must be taken into account in the network modelling process. Especially
the channel characterization, and interference control mechanisms in the case of
any CDMA system must be considered. In W-CDMA network multiple services coexist. Different services (voice, data) have different processing gains, Eb/No
performance and thus different receiver SNR requirements. In addition to these
the W-CDMA coverage depends on the load characterization, hand over
parameterization, and power control effects. In current second generation
systems' coverage planning processes the base station sensitivity is constant and
the coverage threshold is the same for each base station. In the case of W-CDMA
the coverage threshold is dependent on the number of users and used bit rates in
all cells, thus it is cell and service specific.
The W-CDMA planning process can be divided into three phases: initial
planning (dimensioning), detailed radio network planning and network
operation and optimization. Each of these phases requires additional support
functions like propagation measurements, Key Performance Indicator
definitions etc. In a cellular system where all the air interface connections
operate on the same carrier the number of simultaneous users is directly
influential on the receivers' noise floors. Therefore, in the case of UMTS the
planning phases cannot be separated into coverage and capacity planning. In the
case of the post-second generation systems data services start to play an
important role. The variety of services requires the whole planning process to
overcome a set of modifications. One of the modifications is related to the
quality of service (QoS) requirements. So far it has been adequate to specify the
speech coverage and blocking probability only. Also one has to consider more
and more the indoor and in-car coverage probabilities. In the case of UMTS the
problem is slightly more multi-dimensional. For each service the QoS targets
have to be set and naturally also met. In practice this means that the tightest
requirement shall determine the site density. In addition to the coverage
probability the packet data QoS criteria are related to acceptable delays and
throughput. Estimation of the delays in the planning phase requires good
knowledge of the user behaviour and understanding of the functions of the
packet scheduler.
Common features between second and third generation coverage prediction
also exist. In all the systems both of the links have to be analysed. In current
systems the links tend to be in balance whereas in the case of third generation one
of the links can be higher loaded than the other, and thus either one of the links
could be limiting the cell capacity or coverage. The propagation calculation is
basically the same for all standards, with the exception that different propagation
models could be used. Another common feature is the interference analysis. In the
case of W-CDMA this is needed for the loading and sensitivity analysis, in the case
of TDMA/FDMA it is essential for frequency allocation. In order to fully utilize the
W-CDMA capabilities, a thorough understanding of the W-CDMA air interface is
needed from the physical layer to the network modelling, planning and
performance optimization.
In this paper the pre-operational phase of the W-CDMA planning process, as
depicted in Figure 1 in detail, is discussed. Section II concentrates on the initial
planning issues. The W-CDMA link budget is introduced and it is demonstrated
how different services and their QoS requirements impact on the site density
estimate. In Section III a static radio network simulator is introduced. The
Figure 1.
W-CDMA radio
11.1
W-CDMA
In this section the W-CDMA uplink and downlink budgets are discussed. To
estimate the maximum range of a cell a RLB calculation is needed. In the RLB the
antenna gains, cable losses, diversity gains, fading margins, diversity gains etc.
are taken into account. The output of the RLB calculation is the maximum
allowed propagation path loss which in return determines the cell range and thus
the number of sites needed. There are a few W-CDMA specific items in the link
budget if compared with the current TDMA based radio access system like GSM.
These include interference degradation margin, fast fading margin, transmit
power increase and soft handover gain.
The interference degradation margin is a function of the cell loading. The
more loading is allowed in the system, the larger interference margin is needed
in uplink, and the smaller is the coverage area. The uplink loading can be derived
as follows, for simplicity the derivation is performed with service activity v = 1.
To find out the required uplink transmitted and received signal power for a
mobile station MSk connected to a particular base station BSn, the basic CDMA
Eb/No equation is used. The usual, slightly theoretical, assumption is that I oth, the
interference received from the MSs connected to the other cells, is directly
proportional (proportionality constant i) to Iown, the interference received from
the MSs connected to the same BSn as the desired MS. Assume that the MSk uses
bit rate Rk, its Eb/No requirement is k and the CDMA modulation bandwidth is W.
Then the received power of the k-th mobile, pk, at the base station it is connected
to, must be at least such that
If the equations in (2) are summed over the mobile stations connected to BSn
then
Advances in
UMTS technology
151
where m is the number of services used. The difference between equations (5)
and (6) are due to the fact that (6) does not include sectorization gain and that in
the derivation starting from equation (1) the denominator is I,,,, - p k + do,,, +N
rather than I,,, + il,,,,,, +N, which is only the case when p k << lo,.
The downlink dimensioning is following the same logic as the uplink. For a
selected cell range the total base station transmit power ought to be estimated.
In this estimation the soft handover connections must be included. If the power
is exceeded either the cell range ought to be limited, or number of users in a cell
has to be reduced. For downlink the loading (T&
is estimated based on
where Lpmi is the link loss from the serving BS m to MS i, Lpni is the link
loss from another B S n, to M S i , pi is the transmit E,/N, requirement for the
MS i, including the SHO combining gain and the average power raise caused
by fast power control, N is the number of base stations, I is the number of
connections in a sector and ai is the orthogonality factor depending on
multi-path conditions (a= 1: fully orthogonal).
The term
Lpmi
Direct output of the downlink RLB is the single link power required by a user
at the cell edge. The total base station power estimation
must take into account
multiple communication links with average (ZPmi)
distance from the serving
base station. Furthermore, the multi-cell environment with orthogonalities ai
should be included in the modelling. More on the downlink loading and transmit
power estimations can be found in [13]. In the RLB calculation in uplink direction
the limiting factor is the mobile station transmit power, in downlink direction the
limit is the total base station transmit power. When balancing the uplink and
downlink service areas both links must be considered.
The interference degradation margin to be taken into account in the link
budget due to a certain loading q (either in uplink or downlink) is
Fast fading margin or power control headroom is another CDMA specific item
in the RLB. Some margin is needed in the mobile station transmission power for
Soft handover gain is discussed already in [3]. Handovers - soft or hard provide gain against shadow fading by reducing the required fading margin. Due
to the fact that the slow fading is partly uncorrelated between cells, and by
making handovers the mobile can select a better communication link.
Furthermore, soft handover (macro diversity) gives an additional gain against
fast fading by reducing the required Eb/N0 relative to a single radio link. The
amount of gain is a function of mobile speed, diversity combining algorithm
used in the receiver and channel delay profile. More about SHO gain can be found
in Section II.3 .
where
where Nf, is the receiver noise figure, is the Boltzmann constant, T0 is the
absolute temperature and is the loading. In some cases the basic
noise/interference level is further corrected with the man-made noise term.
where
where Pout is the outage at the cell edge, SHO is the fading margin in the case of
soft handover, a is the standard deviation of the field strength and for 50%
correlation a = b = 1/ 2. With the theory presented for example in [1] this
probability at the cell edge can be converted to the area probability. In the
W-CDMA link budget the SHO gain is needed. The gain consists of two parts:
combining gain against fast fading and gain against slow fading. The gam
against slow fading is dominating and it is specified as:
After choosing the cell range the coverage area can be calculated. The
coverage area for one cell in hexagonal configuration can be estimated with:
where S is the coverage area, r is the maximum cell range and K is a constant,
depending on the network topology. The number of sectors is typically from
1 to 3. In the case of W-CDMA reasonable values are up to 6 sectors. In the case
of 6 sectors the estimation of the cell coverage area becomes problematic, since
UL
Transmitter power
Tx antenna gain
Cable/body loss
Transmitter EIRP
(incl. losses)
Thermal noise density
Receiver noise figure
Receiver noise density
Receiver noise power
Interference margin
Required Ec/Io
125.00
20.97
0.00
2.00
18.97
-174.00
5.00
-169.00
-103.13
-3.01
-17.12
mW
dBm
dBi
dB
dBm
a
b = 10 log10(a)
c
d
e = b + c-d
1372.97
31.38
18.00
2.00
47.38
- 174.00 dBm/Hz
dB
dBm/Hz
dBm
dB
dB
8.00
g
- 166.00
h = f+g
i = 101og10(W)+h -100.13
- 10.09
j
-7.71
k = 10 log10(Eb
/N 0 /(W/R))-j
- 107.85
Required signal power [S] -120.26 l = i + k
0.00
18.00 m
Rx antenna gain
2.00
2.00
n
Cable/body loss
95.00
95.00
Coverage probability
(outdoor requirement)
0.00
0.00
Coverage probability
indoor (requirement)
85.62
85.62
Outdoor location
probability (calculated)
32.33
32.33
Indoor location
probability (calculated)
outdoor
outdoor
Limiting environment:
7.00
7.00
Log normal fade
constant outdoor
12.00
12.00
Log normal fade
constant indoor
3.50
3.50
Propagation model
exponent
-7.27
0
-7.27
Log normal fade margin
2.00
0.00
Handover gain
P
(inc. any macro
diversity combining
gain at cell edge)
-5.27
-7.27
Slow fade margin
q =o+p
0.00
r
0.00
Indoor loss
0.00
s
0.00
TPC headroom
(fast fade margin)
147.96
147.96 t = e- l + mAllowed propagation loss
n + q + r- s
dBm
dBi
dB
%
%
%
%
dB
dB
dB
dB
dB
dB
dB
dB
a six-sectored site does not necessary resemble a hexagon. A proposal for the cell
area calculation at this stage is that the equation for the omni case is used also in
the case of 6 sectors and the larger area is due to a higher antenna gain. The more
sectors are used the more careful soft handover overhead has to be analysed to
provide an accurate estimate. In Table II some of the K-values are listed.
1 omni
2-sectored
3 -sectored
6-sectored
2.6
1.3
1.95
2.6
area coverage analyses for UL and DL dedicated channel as well as for common
channels (common pilot CPICH, broadcast control channel BCCH, forward access
and paging channel FACH and PCH on the P-CCPCH and S-CCPCH) can be
performed. In case a second carrier is present in the network area, either used by
the same operator or by another operator adjacent channel interference (ACI) can
be taken into account. Only when the second carrier is assigned to the same
operator, load can be shared according different strategies between the carriers
(IF-HO).
where Pk is the total transmit power of the base station to which link k is
established, Lpkm is the link loss from the cell k to the mobile station m, k is the
cell specific orthogonality factor, pkm is the power allocated to the link from base
station k to mobile station m, Ioth, m is the other cell interference and Nm is the
background and receiver noise of MS m.
The initial transmit powers are adjusted iteratively according to the
difference between the achieved and the targeted C/I value until convergence is
achieved. The process requires iteration, since the C/I at each mobile station is
dependent on all the powers allocated to the other mobile stations and it is not
known a priori whether a link can be established or not. In case either certain
link power limits or the total transmit power of a base station is exceeded mobile
stations are performing IF-HO if allowed or taken out randomly from the network.
In a further step for each mobile station it is checked whether the received
Ec/I0 value is above a user defined threshold so that the mobile station can
reliably measure the base station and synchronize to it. Also, here if the threshold
given is exceeded, the mobile station tries IF-HO or is put to outage. A flow chart
for the detailed iteration steps can be seen in Figure 3.
Figure 3.
Flowcharts for the UL and DL iteration steps.
Advances in
UMTS
technology 161
Figure 5.
DL
where variables are as defined after equation (19), Ioth is interference from other
cells on same carrier and I,, is adjacent channel interference.
where P CPiCH is the CPICH power of the best server, Lp is the link loss to the
best server, PtX i, is the total transmit power of BS i, Lpi is the link loss to BS
i, I ACI is adjacent channel interference, N0 is the thermal noise of the default
MS and numBSs is the number of base stations in the network. The achieved
Ec/I0 is then compared to a user given threshold and the CPICH coverage
defined as the ratio of pixels where the threshold is exceeded compared to all
pixels. The weakness of this modelling for the CPICH EC/I0 coverage is that it is
done only for the best server. In an operating network however, all neighbour
cells must also be measured and therefore all neighbour cells' CPICH EC/I0
should be analysed, too. This could be overcome however by e.g. adding a
threshold to the required CPICH Ec/IQ. Descriptions of other static simulators
can be found for example in [18, 19].
available from the operator. In the second phase the network is planned for the
estimated site distance (1.5*R) and the W-CDMA analysis is performed for the
radio network. As depicted in Figure 1 this case is concentrating on the first
half of the radio network planning process: the network dimensioning,
network configuration definition and coverage/capacity planning.
In the first phase the dimensioning task is to generate certain traffic and QoS
requirements for the above mentioned three cases and to dimension the cases
accordingly. The radio network was dimensioned for an urban area of
13 X 13 km2 . The propagation model used was Okumura-Hata with an area
correction factor of 0 dB. The traffic and QoS requirements for all the three cases
is shown in Table III.
Subscribers
60 000
Average
traffic/subscriber
Blocking
Coverage probability
Data activity
Assumed soft capacity
SHO overhead
Maximum uplink
loading
MS/BS transmit power
Common channel
overhead
Antenna gain MS/BS
Cable/body loss
MS speed
0.060 Erl
2%
95%
50%
20%
40%
60%
SPEECH
AND 64
KBPS CS
DATA
60 000,
12 000
0.040 Erl ,
0.025 Erl
144 KBPS
PACKET
DATA
1%
80%
95%/70%
24 000
3 kbps
50%
10%
40%
65%
10%
40%
65%
24/43 dBm
24/43 dBm
24/43 dBm
10%
10%
10%
0/17 dBi
3/0 dB
3km/h
120 km/h
0/17 dBi
3/0 dB
50 km/h,
3 km/h
0/17 dBi
3/0 dB
20 km/h
7 dB
7 dB
7 dB
4 dB
4 dB
4 dB
50%
55%
30 m
50%
55%
35 m
50%
55%
100%
35 m
In the dimensioning process both the capacity and the coverage have to be
considered. The cell range is determined not only by the RLb, but also on the
capacity requirements. From the selected cell range the coverage area for the
site can be estimated by equation (17). Once the site coverage area is known
the supported traffic density can be estimated. The selected site density must
be such that the traffic density requirement and the coverage requirement in
terms of coverage probability are met. In the following table the dimensioning
results are collected. In the example case the cell range has been solely based
on the uplink results. In the uplink direction the target loading as specified by
equation (6) is limiting the number of channel elements per cell. As it can be
seen already in the dimensioning results, the base station transmit power is
actually limiting in all of the cases. In order to meet the capacity requirements
of either cell range, the number of sectors or the number of carriers should be
changed. In this case only one carrier was used.
In the second phase of the work the dimensioning results in terms of number
of cells and the cell range were used in the corresponding static simulator runs.
The items with bold font in Tables III and IV were used as inputs for the
simulation. The site distance used in the static simulator has been estimated
with D = 1.5*R where R is the cell range obtained from the RLB in
dimensioning. In this study the network scenario was based on a regular grid,
and the network area consisted only of urban area type. The static simulator
Table TV. Dimensioning results.
SPEECH
SPEECH
AND 64
KBPS CS
DATA
144 KBPS
PACKET
DATA
2.08 km
1.98 km
1.52km
Uplink loading
0.32 3 km/h
0.29 120 km/h
0.40 speech
0.18 data
0.65 data
Service limiting
the cell range
Speech 3 km/h
64 kbps data
42 speech
6 data
855.6 kbps
throughput ~
6 data
channels
Number of channel
elements per cell
supported by DL
21
3 km/h
19 120 km/h
27 speech
6 data
5 data
Required number
of 3-sectored sites
(based on UL)
20 (60 cells)
22 (67 cells)
37 (112 cells)
described in Section 111 was used in this study. The %/No requirements and the
traffic distribution used in the simulation were the same as were used during
dimensioning. In the dimensioning phase only the antenna gain is used. In the
simulations the antenna radiation pattern is also required and an antenna with
65" horizontal beamwidth was selected. The 65" antenna was selected because
it is widely used already in 2G networks with three sectored configurations.
Furthermore, according to [ 111 the 65" provides the optimum performance for
the used network configuration. In all the simulations the antenna gain was
always 17 dBi. The soft handover addition window was set to - 6 dB. In Figure
6 there is an example of the network scenario. The system features used in the
simulations are from [ 101 and the ITU vehicular A channel [5] has been assumed
for the channel delay profile.
simulation run so that the ratio of the data users and the speech users would be
exactly as required. The initial requirement was that the data users are 17% of
the total users. After the static simulator run the percentage was 16%. In the case
of UL there are on average 37 speech users instead of 42, but on the contrary
there are 8 data users instead of 6. Therefore one can claim that the uplink
dimensioning result is well in line with the simulation results. The power
amplifier dimensioning for the DL (based on [13]) shows a clear downlink
limitation. The uplink 48 users cannot be served in downlink direction with the
assigned 20 W maximum base station transmit power. When the number of users
in the dimensioning was reduced to 27 speech and 6 data users the dimensioning
estimates the total TX power of 17 W including 10% common channel overhead.
Values used for the power calculation were
iDL (=
"
2_,
Pmi
in
SPEECH
AND 64
KBPS CS
DATA
144 KBPS
PACKET
DATA
37 speech
8 data
25 speech
5 data
97.6%
74.4%
0.62
0.75
0.67
18.95
0.62
19.10
87.5%
For the mixed traffic (speech and 64 kbps) case the probabilities are 98%
and 70% respectively. Since data services are limiting the range the actual
coverage probability for speech was higher, i.e. 98%. Similar analysis for the
RT data gave 68% coverage probability. In the dimensioning the used
requirement was 70%.
The results presented here demonstrate that in macrocellular environment the
downlink dimensioning is essential, due to the fact that with the assumptions of
the macrocellular environment and Vehicular A channel conditions the network
is downlink limited.
Uplink dimensioning alone gives too optimistic capacity predictions for the
network. Furthermore, the modelling proposed for the dimensioning is verified
with static network simulations. Comparison of the simulation results and the
dimensioning results shows good agreement.
Figure 7. Example result of the mixed
traffic case. The coverage probability for
speech services.
where N is the total number of active mobile stations in the system and m is
index for the observed user. Lpn, k is path loss (attenuation due to distance and
slow fading) between the base station k, and the mobile station n. g/g
multi-path fading normalized to having long term average equal to one and J is
the number of multi-path components. pms(n) is the transmission power of the
mobile n. After the interference calculations, the uplink signal-to-noise ratio
SNRUL can be calculated for the user m connected to the base station k as
where Ims(m) is the total interference power received by the mobile station m, M
is number of base stations in the active set, pbs(m,k) is the transmitting power for
the observed user from the base station k, Pbs(k) is the total power transmitted
from the base station k, ak,,i is amplitude attenuation of the channel tap i and Jk
is the number of allocated RAKE fingers from base station k.
In the dynamic simulator the following items were measured: bad quality
calls, defined as calls having an average frame error rate FER exceeding a
threshold (usually 5% for speech). The minimum call duration is set to 7 sec in
order to increase the confidence of the averaging. Statistical data of these calls
such as coordinates, start and end time and the call duration are recorded .
Dropped calls, i.e. calls that have consecutive frame errors that exceed a
threshold (usually 50 frame errors). Usually dropped calls are considered as
severely poor quality calls. So bad quality and dropped calls can be taken as one
measure whose percentage is referred to the number of started calls after the
warm-up period. Power outage - for speech, this is taken from active terminals
including those that are in DTX. Therefore it is slightly distorted due to the other
half of the users that are in DTX. So the actual outage for terminals that are
actively "talking" is higher, rough value is twice than that of the output. There is
no discrepancy for data. E b /N 0 targets are taken from all active terminals
including those in SHO. So all factors regarding the channel and diversity are
taken into account. Finally a SHO probability histogram of the number of
branches per user was collected.
From the static simulator for all UL and DL connections the histogram of th
transmit powers and their cumulative distribution function are taken. Moreover
the p-th percentiles Qp for 0, 50, 75, 90, 95 and 100% are extracted. Statistics
showing the number and type of SHO connection were gathered. For the whole
simulated area the estimated active set size (AS size) is collected (based on the
CPICH levels.) In each simulated case also the UL loading level was stored.
In the final comparison the total traffic per cell, UL power distribution [dBm],
DL total/link power distribution [dBm], SHO statistics, SHO areas, cell dominan
areas, and not served mobiles (static) versus dropped and bad quality calls
(dynamic) were of interest.
Figure 8 the difference of the cell dominance areas as seen by the two
different simulators is depicted. It can be stated that the difference is minor. In
90-95% of all the pixels both simulators propose the same dominant cell.
The uplink power distribution statistics are collected for the speech and
data case in Table VI. The maximum values do not differ significantly, but
some of the percentile-values are well apart. This could indicate different
power distribution shapes for the two simulators. This cannot be avoided due
to the different nature of the simulators.
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Q50
Q75
Speech case
-44
-10.38 -1.37
-7
-14.5
-49
-5
-4.12
-5.63
Data case
-41.79 -0.08
7.1
-44
-3
6
-2.21
-2.92
-1.1
Q90
Q95
MAX
5.81
0
-5.81
10.95
4.5
-6.45
20.39
20
-0.39
13.59
15
1.41
15.77
19
3.23
20.03
20
-0.03
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Static s.[dBm]
Dyn. s. [dBm]
diff [dB]
Q50
Q75
Speech case
16.12
8.25
13.58
-1
-7
12.5
-9.25
-1.08
-0.38
Data case
18.91 -24.01 24.8
7
25
25.5
0.7
-11.91
0.99
Q90
Q95
MAX
18.37
20
-1.63
18.98
21.5
2.52
24.14
24
-0.14
25.57 25.83
25.7 25.8
0.13 -0.03
26.29
26
-0.29
SHO
1-way
2-way
3-way
OVERHEAD
83.7
14.9
1.4
17.7
72
23
33
Figure 10. Link power distributions. Left figure for the static simulator
(ave Ueta = average UL loading).
Figure. 11.
AS size difference
in speech case.
VI. Conclusion
In this work the radio network planning process and methods for W-CDMA
networks were introduced and verified. Accuracy of tools is essential to provide
an operator with reasonable information of the required network topology and
hardware requirements. In this work initial planning phase (dimensioning)
methods were introduced. Furthermore the dimensioning results were compared
with results proposed by static simulator. The comparison of the static
simulation results and the dimensioning results show good agreement. In general
it can be stated that the proposed dimensioning methods perform with
reasonable accuracy in cases where the traffic distribution is such that the
network can serve several simultaneous users in each cell. In the case of high bit
rate services the network performance is strongly dependent on the location of
the mobile users and the general interference situation, therefore it can be stated
that also the proposed dimensioning modelling will perform with degraded
accuracy. In the case of high bit rate services with low number of connections
the proposed methods would underestimate the required BS power too much
giving only a long-term average. One way to overcome this problem would be
to use the worst case parameter values (cell edge) for the most demanding
services.
Further, it has been demonstrated that static radio network planning
simulator is giving a realization of the network, which is close to the network
analysed by the fully dynamic simulator. Generally the tools result in a similar
picture of the network. The uplink and downlink power distributions as well as
the cell loading levels and supported links per cell are following the same trends.
Problem zones in the network (dropped/bad quality calls, outage etc.) occur in
the same locations in both analyses. Handover probabilities, active set sizes and
dominance areas are almost the same in both tools. Nevertheless, a static tool is
suitable for network planning. A dynamic simulator however is superior for
benchmarking of radio resource management algorithms and for analysing the
dynamic phenomena in the networks. The reason is not only in the computation
complexity of the dynamic one but also in the fact that the dynamic tool is "not
calibrated", i.e. that the call drop rate is dependent on dropping criteria and thus
the number cannot be taken as exact absolute value. The number could just be
compared with different simulation with the same dropping criterion.
Acknowledgements
The authors would like to thank all their colleagues, especially Mr. Kari Sipil,
Dr. Ari Hmlinen and Dr. Toms Novosad for support and valuable
contributions.
REFERENCES
[1] JAKES (W.C.), Microwave mobile communications, IEEE PRESS,
p. 126.
[2] REUNANEN (J.), "Multiple Server Location Probability in GSM/DCS1800 Cellular
Systems", Master of Science in Engineering Thesis, Helsinki University of
Technology, (1997).
[3] VITERBI (A.J.), CDMA Principles of spread spectrum communication,
Addison-Wesley, p. 198, (1995).
[4] HATA (M.), "Empirical Formula for Propagation Loss in Land Mobile Radio
Services", IEEE Transactions on Vehicular Technology, VT-29, n 3, (Aug. 1980).
[5] Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000,
Recommendation ITU-R M. 1225, (1997).
[6] SIPIL (K.), LAIHO-STEFFENS (J.), JSBERG (M.), WACKER (A.), "Modeling the Impact of
the Fast Power Control on the WCDMA Uplink", Vehicular Technology Conference,
Houston, (May 1999).
[7] SIPIL (K.), JSBERG (M.), LAIHO-STEFFENS (J.), WACKER (A.), "Soft Hand Over Gains in
Fast Power Controlled WCDMA Uplink", Vehicular Technology Conference,
Houston, (May 1999).
[8] TOSKALA (A.), HOLMA (H.), MUSZYNSKI (P.), "ETSI WCDMA for UMTS", Proceedings of
ISSSTA98, South Africa, (Sept. 1998).
[9] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), JSBERG (M.), "Static Simulator for
Studying WCDMA Radio Network Planning Issues", VTC99, Houston.
[10] http://www.3gpp.org/.
[11] WACKER (A.), LAIHO-STEFFENS (J.), SIPIL (K.), HEISKA (K.), "The Impact of the Base
Station Sectorisation on WCDMA Radio Network Performance", VTC99
Amsterdam.
[12] A. SAMPATH (A.) et a/., "Erlang Capacity of a Power Controlled Integrated Voice
and Data CDMA System", IEEE 47th Vehicular Technology Conference Proceedings,
3, Phoenix, Arizona, pp. 1557-1561, (May 1997)
[13] SIPIL (K.), HONKASALO (Z.), LAIHO-STEFFENS (J.), WACKER (A.), "Estimation of Capacity
and Required Transmission Power of WCDMA Downlink Based on a Downlink Pole
Equation", in Proc. of the IEEE VTC2000-Spring conf., Tokyo (May 2000).
[14] LAIHO-STEFFENS (J.), SIPIL (J), WACKER (A.), "Verification of 3G Radio Network
Dimensioning Rules with Static Network Simulations", in Proc. of the IEEE
VTC2000-Spring conf., Tokyo, (May 2000).
[15] HMLINEN (S.), HOLMA (H.), SIPIL (K.), "Advanced WCDMA Radio Network
Simulator", Proceedings of PIMRC 1999, Aalborg, Denmark, pp. 509-604,
(Oct. 1997).
SALONAHO (O.), "A Novel Interface Between Link and System Level Simulations",
Proceedings of ACTS Summit 1997, Aalborg, Denmark, pp. 509-604, (Oct. 1997).
[17] BRADY (P.T.), "A Model for Generating On-Off Speech Patterns in Two-Way
Conversation", The Bell Systems Technical Journal, 48, n 7, (Sept. 1969).
[18] DEHGHAN (S.), LISTER (D.), OWEN (R.), JONES (P.), "WCDMA capacity and planning
issues", IEE Electronics & Communication Engineering Journal, pp 101-118,
(June 2000).
[19] LABEDZ (G.), LOVE (R.), "A New Time-based Outage Criterion for the Forward and
Reverse Links of DS-CDMA Cellular Systems",VTC98,Toronto, pp. 2182-2186.
Chapter 7
An open software-radio
architecture supporting
advanced 3G+ systems
C. Bonnet, G. Caire, P. Humblet, A. Nordio
D. Nussbaum and R. Knopp
Institut Eurcom, France
A. Enout
UDcast, France
G. Montalbano
Philips Semiconductors, France
T. Hhne
Nokia Research Center, Helsinki, Finland
B. Rimoldi
Mobile Communications Laboratory, Switzerland
users, and eventually detecting the transmitted symbols. All these operations
are performed in real-time on our platform.
The adopted modulation scheme (BPSK and QPSK), the spread signal
bandwidth (5 MHz), the spreading gain (DS-CDMA with allowed spreading gain
of 1, 2, 4, 8, 16) and the frequency band (around 2.1 GHz) are the same as those
defined in the UMTS/TDD specifications. The TDD mode has been chosen because
under certain circumstances it allows the exploitation of the channel reciprocity
between uplink and downlink in duplex communication, and also reduces the
DSP computational load for a given bandwidth.
The solutions proposed in this paper can also be applied to a wide class
of linearly-modulated digital signals (including most of today's and future
mobile communications standards, like GSM, IS-54, IS-136, IS-95, and DECT
(see e.g. [6] and references therein), UMTS (both FDD and TDD modes) [7],
CDMA2000 (see e.g. [8, 9]) and EDGE (see e.g. [10] and references therein).
We show a simplified block diagram of the entire system for a single antenna
in Figure 1.
We have considered two software implementations, the first using a
combination of commercially available embedded DSP cards and a common PC
and the second using the DSP units (e.g. MMX) of a standard PC under an operating
system proving hard real-time support (e.g. RTLinux). Both run in real-time and
are compatible over the air.
II. 1. RF front-end
A simplified block diagram of our RF front-end is shown in Figure 2. It was
designed in conjunction with STMicroelectronics in Geneva, Switzerland. On
reception (Rx), the RF signal is filtered, amplified and down-converted to a 70
MHz intermediate frequency (IF). The local oscillator frequency is digitally
tunable in steps of 500 Hz. The IF signal is amplified by a digitally tunable gain
control and directly sampled at 14.7456 MSamp/sec. The samples are transferred
via a ribbon cable using high-speed line drivers to the data acquisition unit
(DAQ). The mode of the transceiver (i.e. transmission, reception, calibration) is
fully controllable from the software portion of the platform.
On the transmission end (Tx), samples from the DAQ arrive at a rate of
14.7456 MSamp/sec and drive a hardware up-sampling circuit and a
high-speed D/A converter (117.9648 MSamp/sec) to directly synthesize the 70
Figure 2. RF front-end.
MHz IF signal. This procedure is described in more detail in Section II. 1.2.
This signal is then amplified, up-converted to RF, filtered and amplified by a
variable-gain power amplifier.
Special low-speed lines control the gains of the Rx variable-gain amplifier
and Tx Power Amplifier, LO frequency, as well as the antenna switch. Although
not included in Figure 2, automatic wide-band calibration capabilities (for both
Tx and Rx) are included in the design, which are required for multiantenna
systems (see e.g. [11]). These will be used in a later stage of the project where
the architecture will be extended to implement multiple antenna transceivers.
The basic characteristics of the RF front-end are summarized in Table I.
II.l.l. Passband A/D conversion
Once the RF signal at the antenna has been down-converted to IF, it is sampled
by an A/D converter at a certain rate fADC (Figure 9). Calling r IF (t) the received
IF analog signal and choosing fADC >2W according to [4], because of the
periodicity of the discrete-time signal spectrum, the resulting real sampled signal
r[n] = r I F (n/f A D C ) is pass-band with a spectrum replica centred at fADC/4
(although fIF and fADC at the receiver can be different from fIF and fADC at the
transmitter, for simplicity we use the same notation).
Notice that here in order to avoid signal re-sampling we suppose the rate fADC
to be a multiple integer of the chip rate (i.e. fADC = Ncfc where in our
implementation we set Nc = 4).
Although a base-band version of the received signal can be obtained by
multiplying r[n] by (-j)n and then by low pass filtering, we will show further
how the channel estimation and the data detection processes can be performed
at pass-band.
182 An
and by hD/A[n] the filter impulse response. The filter output is given by
If the filter impulse response has length LD/A, there is only a single
non-zero term in the sum in the right-hand side of the above equation. Then,
x"'[n] = hD/A[m]x'[k], where k= [n/LD/A] and m = n modulo LD/A. Therefore,
the computational cost of the filtering operation consists of one real product
per output sample at rate fd. After the D/A conversion, the continuous-time
signal y(t) is eventually up-converted to the RF carrier and sent to the antenna.
In this way, the periodic spectrum of x'[n] shows a spectrum replica centred
at fIF (see Figure 7(a)). After D/A conversion, a pass-band filter centred at fIF
removes the other replicas, generating the desired IF modulated signal.
The discrete-time modulation by fDAC/4 in (5) requires a negligible
computational cost since it corresponds to change alternatively the signs of x[n]
as can be noticed in expanding equation (5). In order to avoid aliasing when
taking the real and imaginary part, the sampling rate must satisfy also the
condition fDAC >2W.
III.3. Receiver
In this section we analyse some of the theoretical aspects of the receiver
signal-processing. In particular we give a description of the receiver front-end
architecture shown in Figure 10. Then we focus on channel estimation, matched
filter synthesis and symbol detection.
III.3.1. Frame synchronization
Frame synchronization is achieved using a filter matched to the primary
synchronization sequence to estimate the location of the start of a frame. This is
achieved by filtering the bandpass received signal r[n] as
(a) Spectrum of x[n], x'[n], and y(t) with the integer l = 2 and the sign (+) chosen in (4)
and averaging the rs[n]2 over several frames. The maximum output of this filter
is used to adjust the receive signal strength (via a variable gain IF amplifier) and
synchronization is achieved when the maximum is greater than a pre-defined
threshold. Note that this filtering operation involves purely real quantities. This
is typically the most computationally intensive part of the receiver front-end
since it requires a fairly long filter operating at the sampling rate. The 3GPP
standard uses a hierarchical structure for the primary synchronization sequence
which allows the filter to be implemented as a concatenation of 2 FIR filters of
length 16. A block diagram of the primary synchronization sequence is shown in
Figure 11.
where
is the u-th user channel filter vector and v is a vector of interference plus noise
samples, assumed to be white. The MNC x MNC matrix A is defined as
and where a is the Discrete Fourier Transform (DFT) of a. After some algebra
[18], it is possible to show that the Least Squares estimation of the overall
channel impulse response g is given by
where DFT and IDFT denote direct and inverse Discrete Fourier Transforms. The
ratio of two vectors should be interpreted as the element-by-element division.
This approach can be applied to both base-band and pass-band signals. The
receiver can also use the a priori information that the signal bandwidth is limited
to W. Notice that this operation in the frequency domain corresponds to low-pass
filtering in the time domain, moreover it reduces the computational cost
since only a part of the MNC products (by the element-wise inverses of in (12))
are computed. Eventually, after the IDFT, the processing gives the estimated
where Li is again the spreading gain. Using the sampled channel estimate g[k]
we synthesize the discretetime filter fi[k] = fi(t) | t=k/fADC matched to the overall
response as follows
In order to extract the data symbols we filter the received signal with the MF
obtaining
In this setting r[k] is real while fi[k] is complex so the product requires two
real multiplications. To the contrary the baseband samples of r[k] would be at
half the sampling rate but would be complex. So the two complexities are
identical.
Notice that the signal after matched filtering is still pass-band and the symbol
estimates after sub-sampling are given by
In this way the symbol estimates are given by sub-sampling the MF output at
symbol rate without taking care of the demodulation.
III.3.4. Carrier synchronization and decoding
The carrier synchronization is done at symbol rate with a classical
decision directed algorithm [20]. The algorithm then takes a decision on
the symbols and recovers the data (in our example a video stream).
III.3.5. Re-sampling
The baseband processing algorithms such as synchronization, channel
estimation and data detection assume that the signal is sampled with an integer
number, Nc, of samples per chip. In the current implementation we set NC = 4.
This solution avoids utilization of re-sampling techniques at both the
transmitter and receiver front-end. These techniques have been studied in [15]
and will be implemented in the next version of the software.
V. Conclusion
This first demonstration shows that the architecture of the platform is capable
of sustaining real-time communications and is thus promising for future
developments. The platform is currently being enhanced and opened to both
industrial and academic collaboration. The enhancements will consist of:
support for multiple antenna transceivers;
more sophisticated signal processing algorithms;
multi-user detection;
layer 2 (RLC,MAC) functionality.
Collaboration has already begun under the label of the RNRT (Rseau
National de la Recherche en Telecommunications) financing programme
organized by the French Ministry of Industry and Finance. Three projects have
been initiated covering the following topics:
radio sub-system improvement (flexibility and sensitivity);
compliance with the3GPPUMTS/TDD specification;
higher-level protocol stacks;
integration to an IPv6 experimental backbone.
Acknowledgements
The authors would like to thank D. Margairaz and P. Favrat, formerly with
STMicroelectronics, Geneva, for their collaboration in the development of the RF
subsystem. In addition, the help of G. Vaucher and P. Bruchlmeier from
ACORT/EPFL in the fabrication of many of the hardware components was greatly
appreciated.
REFERENCES
[1] SRIKANTESWARA (S.), REED (J.H.), ATHANAS (P.), BOYLE (R.), "A soft radio architecture
for reconfigurable platforms", IEEE Communications Magazine, (Feb. 2000).
[2] "Special issue on software radio", IEEE JSAC, Vol. 4, (April 1999).
[3] "Software radio", IEEE Personnal Communications, Vol. 4, (Aug. 1999).
[4] TSURUMI (H.), SUZUKI (Y.), "Broadband RF stage architecture for
software-defined radio in handheld terminal applications",
IEEE Communications Magazine, (Feb. 2000).
[5] ZANGI (K.), KOILPILLAIR (R.), Software radio issues in cellular base stations",
IEEE JSAC, Vol. 17, pp. 561-573, (April 1999).
[6] RAPPAPORT (T.), Wireless Communications, Englewood Cliffs, (1996).
[7] 3GPP-TSG-RAN-WG1, "TS-25.2xx series", tech. rep., (Jan. 2000).
[8] DAHLMAN (E.), GUDMUNDSON (B.), NILSSON (M.), SKLD (J.), AB (E.R.S.),
"UMTS/IMT-2000 based wideband CDMA", IEEE Communications Magazine,
Vol. 36, pp. 70-80, (Sept. 1998).
[9] "http://www.itu.int/imt/2-rad-devt/index.html".
[10] FURUSKR (A.), MAZUR (S.), MLLER (F.), OLOFSSON (H.), SYSTEMS (E.R.), "EDGE: Enhanced
data rates for GSM and TDMA/136 evolution" IEEE Personnal Communications,
Vol. 6, pp. 56-66, (June 1999).
[11] MOGENSEN (P.), KREDERIKSEN (F.), DAM (H.), OLESEN (K.), LARSEN (S.), "A Hardware
Testbed for Evaluation of Adaptive Antennas in GSM/UMTS", Vol. IEEE PIMRS'96,
pp. 540-544, (Oct. 1996).
[12] XILINX, "Xilinx 2001 FPGA Databook", http://www.xilinx.com, (2001).
[13] PLXTECHNOLOGIES, "PCI9080", http://www.plxtecj.com, (1999).
[14] 3gpp, "3rd Generation Partnership Project", http://www.3gpp.org, (2001).
[15] MONTALBANO (G.), NORDIO (A.), "Front-end software-radio review",
tech. rep., Institut Eurcom, (July 1999).
[16] STEINER (B.), JUNG (P.), "Optimum and suboptimum channel estimation for
the uplink of cdma mobile radio systems with joint detection",
European Transaction on Communications, Vol. 5, pp. 39-49, (Jan.-Feb. 1994).
[17] 3GPP-TSG-RAN-WG1, "TS-25.221 v3.1.0; physical channels and mapping of transport
channels onto physical channels (TDD)", tech. rep., (Jan. 2000).
[18] CAIRE (G.), MITRA (U.), "Structure multiuser channel estimation for blocksynchronous DS/CDMA", Submitted to IEEE Transaction on Communications,
(July 1999).
[19] GOLUB (G.H.), LOAN (C.F.V.), Matrix Computation, The John Hopkins
University Press, (1996).
[20] PROAKIS (J.G.), Digital Communications, NY: McGraw Hill, 2nd ed., (1989).
[21] "Spectrum Signal Processing", http://www.spectrumsignal.com,
[22] FSMLABS, "RTLinux", http://www.rtlinux.org, (2001).
(2001).
Chapter 8
Wireless communications+ + +
R. Steele
Multiple Access Communications Ltd, Southampton, UK
plug into the same enhanced GSM backbone core network, whose architecture is
displayed in Figure 1. We observe that the conventional GSM network subsystems, such as the BSS, MSC, VLR, HLR, etc, are present, plus elements to support
GPRS. The UMTS terminology for a BSS is a radio network sub-system (RNS), and
for a BS it is a Node B. Figure 1 shows a RNS with interface IUCS for
circuit-switched data from the radio network controller (RNC) (the UMTS version
of a BSC) to the MSC; interface IUPS from the RNC to the packet node SGSN;
interface Iur between RNCS; and interface Iubis between the RNC and the Node B.
It may be expected that with the advent of software radio the BSS and RNC will
be a single entity, as will the BTS and the Node B. We may therefore conclude
that our networks will evolve to handle both packet and circuit-switched data
and with fibre/radio nodes which can adapt to GSM, and UMTS as well as a range
of other radio interfaces. Packet transmissions may eventually be exclusively
preferred over circuit-switched transmissions.
This approach of a core network handling different radio interfaces is not
new. IS-95, now cdmaOne, another global standard, was conceived as a radio
interface, and its original version was a twin of AMPS in a dual-mode mobile. As
is-95 was CDMA and AMPS analogue FM using FDMA, the new digital system had
to work on the IS-41 AMPS backbone network. Nowadays cdmaOne mobiles are
often single mode, operating in both the 800 and 1900 MHz bands. The
evolution of cdmaOne to the 3G system cdma2000 is relatively straightforward.
cdma2000 will be either wideband CDMA spanning a bandwidth of multiples of
1.25 MHz, or it will use multi-carriers, where each carrier has the cdmaOne 1.25
MHz channel occupancy. Thus the lowest state of cdma2000 is cdmaOne (but an
enhanced version).
The expectation of a single global 3G standard has not materialized. Instead
the ITU calls its IMT-2000 a family of 3G systems that will be inter-worked [2, 3].
The terrestrial component has five systems; the two UMTS systems called IMT DS
(direct sequence) and IMT TC (time code); the cdma2000 system called IMT MC
(multi-carrier); IMT sc (single carrier) which is better known as EDGE; and
strangely IMT FT (frequency time) which is the DECT system. There is also the
mobile satellite system (MSS) component to provide global coverage at low
capacity.
Harmonization is in progress to inter-work UMTS and cdma2000, and as EDGE
is on the UMTS road map, then early software radios (which will not include
down-conversion using a DSP) will be able to combine these systems. Further,
MSSS, like Globalstar, are able to inter-work with both GSM and cdmaOne. So the
realization of a single radio interface that can adapt to most cellular systems is
expected during the first decade.
Figure 1. Network architecture for GSM and UMTS and their evolution.
AUC = Authentication centre
BSS = Base station system
EIR = Equipment identity register
GGSN = Gateway PPRS
support node
HLR = Home location register
RNC = Radio network controller
SGSN = Serving GPRS support node
unsightly as they will not be visible, being embedded in building and city
infrastructures. With base stations (BSs) costing very little, they will be deployed
ubiquitously along with cameras, and whether they will be used regularly or not
will not be important. They will be installed for functionality and convenience,
in the same way that lighting and power outlets are used.
Let us pause to consider the implication of myriads of small BSS and their
accompanying MSS radiating low levels of electromagnetic energy over a wide
frequency range and very small coverage areas. The result will be that people
will be exposed to frequent low level electromagnetic fields. This is in contrast
with the current situation of few BSS and MSS transmitting at much higher power.
Given this the expectation is that the average exposure in the future will be less.
Along with the miniaturization and widespread deployment of BSS will go the
installation of physically small switches/routers and databases, although this
may not happen until the second decade. Gone will be the concept of accessing
remote massive switches and databases. Decision making regarding the efficient
use of resources, particularly radio resources, will need to be fast and tailored to
local events. Knowledge of available resources, where the mobiles are currently
located and where they are predicted to be going, QoS agreements relating to the
active mobiles, current link qualities of all links, and so on, will be locally
utilized in deciding what actions to take. Local mobile networks will have
generic similarities with campus networks having private branch exchanges,
although the switches and databases in future mobile networks will be small and
more localized, e.g., a switch/router in an office block or large office. So with
microcells go micro-networks with their small components. Compared to the
present day the equipment effectively disappears in that it will be coffee mug
size.
We mentioned that many BSS will be co-sited with cameras. Tiny in size, the
cameras will be difficult for people to spot their presence, but they will send and
store images of us. While the ubiquitous camera will be an important aid in
decreasing crime, and allow us to observe events at any distance, they could also
be used by governments to ensure that we obey their wishes.
The small radio cells will be used in numerous ways. There will be wireless
local area networks (W-LANS) that will connect home appliances and
entertainment systems and interact with a user's B-LAN. Entertainment will
follow a user around, appliances will be remotely serviced, and complex security
systems will be a growing business. Whenever possible, people will use the fibre
network for accessing the Internet, avoiding the mobile component with its
limited bit rate. Mobile Internet will be for personalized services when nomadic
or on the move. Logging onto the Internet will need to be much faster and
browsing more effective. Users will prefer closed community networks enabling
them to use services, do shopping, etc, with the feeling of security, and with a
quality-of-service they find acceptable. These closed networks will enable
employees to work at home, or elsewhere, creating secure virtual businesses.
Private mobile radio (PMR) has been used by special groups, such as the
emergency services, police, and transport companies, for many years. PMR was
the precursor to cellular radio. It lingers on into the age of cellular with
insufficient spectrum and without the huge market size of cellular that forces
down the costs of mobile stations and network infrastructure. As cellular
becomes more ubiquitous, inexpensive and with a wide range of services, it will
create virtual PMR networks within its cellular architecture. The result will be that
conventional PMR will cease.
Wireless local loop (WLL) delivering point-to-point services to the customer
via wireless from a BS, has a continuing role to play in developing countries
where installing fibre infrastructure can be difficult and/or expensive. In the
developed world, WLL will linger in the USA, but eventually it will be a blip in
history. It will die because of spreading fibre infrastructure and the wide use of
cellular radio.
While satellite systems have a long term role in monitoring and surveillance,
they have only a medium term existence for communications (as distinct from
TV broadcasting). They will be displaced by aerial platform technology and other
space systems as we will discuss later.
Of the systems existing at the beginning of the 21st century the one that will
last the longest is cellular. It is the king of the wireless world. Yet as time goes
on most cellular communications will not be between people, but between
machines.
Figure 2. Fibre network, different radio cells and an ad hoc B-LAN network.
shifts. Their link budgets are such that line-of-sight radio links are preferred,
and communications from within buildings is not, in general, supported.
Furthermore, satellites are expensive, their launch costs are high, and they
cannot be retrieved from space for servicing and technological updating.
All these disadvantages are tolerated because of global coverage.
MSSS will therefore continue to be used until there is a better alternative to
provide coverage and capacity in areas where it is too expensive to deploy
terrestrial cellular systems. Fortunately there are alternative methods. One that
is actively pursued today are high altitude platforms (HAPS). "High" in this
context means the stratosphere, typically 21 km, and the platforms are either
airships or aeroplanes. The altitude selected is where the stratospheric winds are
minimal and planar. At 21 km the HAPS are above the flying lanes of commercial
aircraft, and they are untethered to earth because of aircraft safety and because
there are no suitable cables. An unmanned airship HAP is required by proponents
to stay on-station for about five years, and convert sufficient solar energy to do
this, as well as provide communications. Antenna arrays attached to the
underside of the airship produce multiple beams that form radio cells on the
surface of the earth. Aircraft HAPS can be either manned or unmanned, fly in
tight circles of some 5 to 8 miles in diameter, and carry arrays that form
terrestrial cells.
HAPS have the following advantages. They can be sited over cities to
increase capacity, or in rural areas where they can provide wide area coverage.
Each HAP can produce tens to hundreds of cells and these cells can be tens of
kilometres to a few hundred metres [8-10]. HAPS are linked to ground stations
and to other HAPS to form HAP sky networks. The cells can be instantly moved
by electronically changing the direction of a beam. Compared to MSSS, the
signal delay of HAPS is small, they do not contribute to the Doppler frequency,
the radio link budget is much better, the spectral efficiency is enormously higher
as the cells are small; they can be returned to earth for technology updates; they
are relatively inexpensive to make and launch, one HAP can produce a cellular
network over hundreds of kilometres, and so on.
The aircraft HAP is the HAP likely to be deployed initially, and as a
consequence, the antenna structures cannot be too large. This means they will
operate at relatively high frequencies (e.g., providing local multi-point
distribution service (LMDS)) or if in the 3G bands, they will only create relatively
large cells. The current strategy is for aircraft HAPS to be piloted, and for crews
to do shifts on-station. This means that for one location, three crews and two to
four aircraft are required. Aerial refuelling of aircraft HAPS facilitating long
flights are not proposed. Flight regulations are not an impediment for aircraft
HAPS.
Airship HAPS offer a better long-term solution (at lower latitudes) as they are
some 200 m in length, can have huge antenna arrays and therefore can create
microcells at 3G frequencies. They must, however, have stabilizing systems,
particularly to compensate for tilting of the antenna arrays, and be held onstation as an operator cannot have a network gently floating away!
Other aerial platforms (APS) will be used. Some will be housed on
extraordinarily high structures, where they will create multiple cells that move
around as required. Others will be on tall towers and tethered balloons. Many
civil aviation aircraft could have APS to form terrestrial cells as they go about
their business of delivering passengers to remote destinations. Some APS will be
small and attached to unmanned aircraft that fly to teletraffic hot-spots, etc. APS
can be at any height, subject to aeronautical constraints. The APS will mark the
demise of MSSS. We might also speculate on the longevity of APS, a technology
that has not yet arrived. With the future ubiquity of fibre networks augmented by
radio tails, the effectiveness of AP technology will decline, but it will always
have the feature of providing many cells at switch-on whose characteristics can
be controlled electronically from the ground.
However, communications from space will continue, based on the large
orbiting space stations that were beginning to be assembled at the end of the 20th
century. We may envisage a network of these space stations, descendents of
MSSS, orbiting the world, of huge size, monitoring earth as well as space, and as
a sideline, providing communications [11]. The problems of Doppler shift and
delay associated with MSS would continue, but the spectral efficiency would be
high as the space stations would form very small cells using very large antennas.
This space station system (SSS), with its orbiting space stations interconnected
via laser beams or high frequency radio links, would be able to deliver mobile
services, very high bit rate point-to-multipoint communications, and broadcast
services. They would also interwork with the aerial platform layers and
terrestrial network beneath them.
laid in many directions the surface platforms connected to them are able to
provide continuous coverage across an ocean. By this arrangement we have
maritime radio cells for ships and aircraft communications connected by a high
speed, high capacity backbone fibre network.
All of these networks, from the international to the one in the refrigerator,
will be soft, i.e., adaptable and flexible to requirements and demands. Such
flexibility, particularly to unforeseen events, requires complex software,
software that is itself changeable. In essence software that, when in a crisis, will
adapt itself. This means code that writes code (we already have this to a limited
extent). These codes will be the software agents (SAS) of the 21st century, the
atom of the software world, the oil that lubricates our interconnecting software
engines.
Let us examine the role of SAs. As an example, consider a B-LAN. When the
wearer speaks, the human interface SA (HI-SA) will, say, visually appear on one
of the user's display systems and ask what duty is requested. Suppose the wearer
is interested in acquiring an antique dressing table. The HI-SA will gather details
for the purchase: the origin of the table, its period, cost, size, when it is needed,
and so on. The HI-SA will have a mobility SA (M-SA) make an access by radio to
a nearby network, visit the yellow pages web site, and obtain a list of service
providers (SPS) who deal in antique dressing tables. The M-SA will then visit the
SAs of the SPs, eventually do a deal, and report the outcome to the user. Suppose
on another occasion the wearer asks the HI-SA what precautions he should take
in climbing a ladder to examine his chimney pot. The HI-SA will ask for
information, and then may need to have a SA devise and use basic equations
involving the theory of statics. If a program is unavailable, the SA will have to
have it written, or ask the M-SA to get either the equations or the result via a SP.
Generally the B-LAN will have the code written locally by body SAs, only
approaching a SP when solving complex problems.
Currently there is much discussion regarding software down-loading to
terminals. Some engineers consider that the terminal should be devoid of much
memory, and consequently the same information will be frequently down-loaded
when required. However, future B-LANS will have formidable storage capacity,
and once acquired, software will be stored locally. This procedure is also in
accordance with people's acquisitive behaviour, e.g. many people prefer to buy
and keep their own books in preference to using a public library.
The network will use SAS to make decisions, and to control, maintain and
adapt their networks. At one extreme we have experienced computer viruses,
which we may think of as malevolent SAS. The SAS created for our networks will
be created by benevolent software robots. They will not, however, be the
equivalent of robots on a car assembly line, programmed to undertake specific
tasks, although they may start off that way. SAS will be large, mobile,
autonomous code and will be created by different engineers with their personal
preferences. They will be conceived to undertake tasks, and inevitably they will
have human-like characteristics. For example, in buying an antique dressing
table the SA will need to be told a lot about the subject as well as handling
business matters. In other words we will need to impart to SAs a lot of human
knowledge, expectations and acceptable goals. Since we also require them to act
independently to unforeseen events that must be dealt with immediately, they
will probably develop genetically. The SAs will therefore occupy a software
world that is psuedo human, and hence SAs will not be good or bad, but have
complex behavioural patterns that will sometimes do things to the networks that
we do not expect.
Although human intelligence evolves extremely slowly, we have been
spectacularly successful in using it to construct modern civilization, and now we
are beginning to understand ourselves. However, we still know relatively little
about our brains and minds. When we consider our network, its processing
power, memory, reconfigurability and evolution are increasing rapidly. When we
apply the word intelligence to the network, or machines in general, it illicits a
howl of protest. This appears to be because intelligence seems to be deemed an
exclusive human attribute, and according to this way of thinking, if a machine
does not have what we think of as human intelligence, it is not intelligent.
You could argue that man has tried to imitate birds, in learning to fly, but has
produced aircraft that cannot flap their wings, dive underwater to catch a fish,
etc. Of course, we did not want to do these things, and we can point out that birds
cannot fly faster than the speed of sound nor carry hundreds of other birds over
thousands of miles. Although man and birds both fly, they do it differently to suit
their respective needs. Similarly, machines and man will both be intelligent, but
in different ways according to their makeup and the goals they need and want to
achieve. So with the passage of time our networks will have highways carrying
enormous quantities of traffic at huge speeds. They will have SAs to allow them
to be self-organizing and reconfigurable. Decisions and executing tasks will be
done at a formidable rate; vast numbers of interconnecting machines will be
accommodated; there will be rising complexity in service provision, and so on.
The SAS will create a software world that will be intelligent.
If this is a glimpse of the future, it follows that we may at this time be
launching a global brain of sorts. This is not like the brain of an individual. It is
more akin to a complex society of interconnecting happenings on many levels;
a representation of our world but in the software domain, the so-called
cyberspace. The hardware of our network, with its multiplexity of wireless and
optical nodes, will somewhat resemble the network's brain, while the SAS will
give it the societal mind.
Industrial strikes by key workers, or picketing of key installations, such as oil
refineries, can bring our societies to a halt in a matter of days. People are unable
to drive their cars to work or to the shops, staff and patients cannot get to their
hospitals, food distribution ceases, and so on. Laws that were fashioned for the
19th and for most of the 20th century seem inadequate to handle the types of
crises that can now suddenly erupt. These flash crises that can bring a country to
a standstill in a few days will be trivial and slow to occur compared to what may
happen in the future when society will be dependent on trillions of
interconnecting processors, SAs, complex ad hoc networks, applications running
at high speed, and so on. If some SAs are forced to make rapid decisions in
response to some unforeseen events, and their decisions result in an almost
instantaneous breakdown of the network, the equivalent of a brain seizure
ensues. Society that will now be so dependent on the brain for almost every type
of decision making, production, distribution and control, will be in a catastrophic
state. In the same way that bulkheads compartmentalize ships so that in the event
of an accident only some parts of the ship will be flooded, we need to design our
networks with software bulkheads. This task should be a major research activity.
A serious fault in the network needs to be rapidly isolated, rather like we isolate
a person with an infectious disease, and dealt with. The "brain" that will make
life for most citizens easy and enjoyable, must be self-healing.
But, will we be able to design and control our global brain with the passage
of time and its growing complexity?
Mankind is faced with a dilemma. As our science is becoming increasingly
complex as time goes by we have fewer people able to understand it. For
example, what proportion of people will be able to comprehend the leading edge
of computer science in fifty years? Very few. Hence we may be inclined either
to do little and arrest mankind's full-ahead mentality, or we go full-ahead
knowing that we risk losing control of our increasingly complex civilization.
There is, however, a way out. Full-ahead to the global brain, but being able to
control it because we biologically change ourselves. We institute changes to
rapidly evolve our own intelligence to keep ahead of our machines. This
involves knowing ourselves in a detailed manner, making biological changes,
and introducing internal body processors so that we can make rapid and complex
decisions. Homo sapiens then forms the genius society and the machines remain
our servants.
Oops, we have gone from 3G to science fiction, or have we?
Acknowledgment
The author thanks the organizers of the ECWT 2000 Conference in Paris, 5-6
October 2000, for permission to amend the paper he presented there, entitled
Full ahead to where?, and to publish it here.
REFERENCES
[1] ZVONAR (Z), JUNG (P.), KAMMERLANDER (K.), "GSM - Evolution Towards Third
Generation Systems", Kluwer Academic Publishers, Boston, (1999).
[2] IEE Conference on 3G Mobile Communication Technologies, London,
(27-29 March 2001).
[3] STEELE (R.), LEE (C.C.), GOULD (P.), "GSM, cdmaOne and 3G systems", John Wiley,
(2001).
[4] STEELE (R.), WILLIAMS (J.E.B.), "Third generation PCN and the intelligent multimode
mobile portable', IEEE Electronics & Communications Eng. J., pp. 147-15,
(June 1993).
[5] STEELE (R.), "Towards a high capacity digital cellular mobile radio system",
lEE Proc. Pt.5, pp. 405-415, (Aug. 1985).
[6] STEELE (R.), "Mobile access technologies - what's next", ACTS Mobile
Communications Summit, Spain, pp. 21-24, (Nov. 27-29, 1996).
[7] HENTSCHEL (T.), HENKER (M.), FETTWEIS (G.), "The digital front-end of software radio
terminals", IEEE Personal Communications, 6, n 4, pp. 40-46, (Aug. 1999).
[8] EL-JABU (B.), STEELE (R.), "Aerial platforms: a promising means of 3G
communications", IEEE Vehicular Technology Conf., Houston, (May 1999).
[9] COLELLA (N.J.), MARTIN (J.N.), AKYILDIZ (I.F.), "The HALO Network",
IEEE Communications, 38, n 6, pp. 142-148, (June 2000).
[10] HASE (Y.), MIVRA (R.), OHMORI (S.), "A novel broadband all-wireless access network
using stratospheric platforms", IEEE vTc '98, pp. 1191-1194.
[11] STEELE (R.), "A vision of wireless communications in the 21ST century",
Shafi (M.), Mizusawa (J.), Ogose (S.), Hattori (T.), "Wireless Communications in
the 21st century', IEEE PRESS, (2000).
Index
3GPP 5
radio interface 6
AAAD 48etseq
AAA-TD 48
access network architecture 8
acquisition indication channel 15
A/D converter 181
adjacent channel interference 160
aliased cross-channels 105
spectral matrix of 105
aliased Tx/Rx global channels 101
asynchronous cell sites 25
three-step search method 25
azimuth-delay 131
spread function 131
CAAAD 48
receiver 48
CAAADIAAA-TD 48
capacity 156
analysis 156
carrier synchronization 193
decoding, and 193
re-sampling 193
CDMA 61 et seq, 98
interpretations 98
models of 98
multicarrier techniques 61 et seq
wideband wireless networks 61 et
seq
cell 154
coverage area estimation 154
range 154
channel 128, 178, 190
estimation 190
flat fading 128
model 128
selection 178
spatio-temporal 129
channelization code 16
code power
interference signal 136
received signal 136
coherent rake combining 30
rake time diversity 30
common pilot channel 14
COMSIC 41
receiver 41
concept 3 (multi-tone - CDMA) 68
coverage 156
analysis 156
CPCH (common packet channel) 22
CPICH coverage 163
214 Index
environments 116
indoor 116
vehicular 116
equal gain combining
technique (EGC) 74
equalizers, ZF and MMSE 109
ETSI 5 , 7
radio interface 7
standardization 5
general architecture 9
UMTS 9
global minimum mean square error
(GMMSE)
detection technique 79 et seq
GSM to UMTS, migration 2
IMT-2000 spectrum 4
indication channel 15
acquisition 15
page I5
indoor 116, 121
with fading 121
interference cancellation 77
parallel 77
successive 77
ITU 3
standardization 3
layer I 19
data format
linear detection
joint 96
linear reception
link 35, 36
forward 36
reverse 35
load system 82
half 86
performance
19
96, 108
92 et seq
82
Index 215
multi-user detection 76, 84
channel coding 84
techniques 76
multi-user transmission 93
network 8, 162
coverage 162
infrastructures 8 et seq
time diversity 30
random channels 117
rate matching 22
multiplexing 22
receiver sensitivity 152
reception 101
re-sampling 193
reverse link 35
RF front-end 180
RT Linux-based PC
implementation 185
scrambling code 17
secondary-common control physical
channel 14
shadowing margin 153
signal 131, 186
digital processing 186
model 131
signal code power 136
interference 136
received 136
signal to interference and noise ratio
(SINR) 108
simulation results 80
detection schemes 80
simulator, dynamic and static 169 et seq
single-user detection 70
analytical results 75
simulation results 75
systems 70
techniques 70,75
SIR 28
measurement 28
fast TPC based 28
site diversity (softkofter handover) 34
smart-antenna 126, 127
soft handover gain estimation 153
softkofter handover 34
software-radio 177 et seq
architecture 177
open 177
space-time 126
processing 133
results 141
UMTS 126
space-time UMTS uplink processing
127 et seq
216 Index
T
TD-CDMA 92
downlink 92
multi-sensor extensions 92
TDD (time division duplex) 92
temporal processing 135
Texas Instruments TMSC6201 184
third generation mobile systems,
UMTS/IMT-2000 1 et seq
standardization 3
third generation partnership project
(3GPP) 5, 7, 12
radio interface 7
standardization 5
three-step search method 25
asynchronous cell sites 25
time division duplex (TDD) 92
time interleaving 85
TPC, SIR measurement-based 28
transmission 99
multi-band 99
transmit diversity 39
transport channel 19,21
multiplexing 19
turbo-code, half load system for 86
U
UL, DCH coverage 162
universal mobile telecommunication
system (UMTS) VII, 2, 9, 126
general architecture 9
and GSM 2
radio access network architecture
(UTRAN) 9
smart-antenna space-time 126
uplink processing 126 et seq
uplink 158
analysis 158
iteration step 158
V
vehicular 116, 119, 120
with fading 120
without fading 119
virtual home environment (VHE) 10
W
W-CDMA 12 et seq, 13, 41, 146, 148 et
seq
capacity enhanced technologies 41
improvements 12
physical channel 13
radio link budget 148
radio network planning process 146
et seq
spreading code assignment 13
wide-band model 104
wide-band wireless networks 61
future of 61
wireless 196
communications 196
Z
ZF 109
equalizers 109, 110